阅读说明：本技术 选择和采收果实的自动化方法和实施该方法的机械设备 (Automated method for selecting and harvesting fruits and mechanical device for implementing same ) 是由 帕特里克·巴尔卢 C·夏巴利耶 于 2021-05-31 设计创作，主要内容包括：本发明涉及果实自动化采收的领域,特别是诸如葡萄之类的成串的果实。本发明涉及一种用于采收茎干所结的果实的方法,该方法包括在两个阶段进行图像处理。根据本发明,方法(100)包括：在表示植物分组的整体图像中检测感兴趣区域的步骤(120),其中使用图像处理过程在整体图像中检测一个或多个感兴趣区域,每个感兴趣区域包括果实的至少一部分,获取局部图像的步骤(130),其中在每个感兴趣区域附近获取局部图像,每个局部图像表示所考虑的感兴趣区域的果实的至少一部分,以及识别茎干的步骤(150),其中在每个局部图像中识别茎干。(The present invention relates to the field of automated harvesting of fruit, in particular bunches of fruit such as grapes. The invention relates to a method for harvesting fruit from stems, comprising image processing in two stages. According to the invention, the method (100) comprises: a step (120) of detecting a region of interest in a global image representing a grouping of plants, wherein one or more regions of interest are detected in the global image using an image processing procedure, each region of interest comprising at least a part of a fruit, a step (130) of acquiring local images, wherein local images are acquired in the vicinity of each region of interest, each local image representing at least a part of a fruit of the considered region of interest, and a step (150) of identifying a stem, wherein a stem is identified in each local image.)

1. A method for harvesting fruit bearing stems or stalks, the method (100) comprising:

a step (120) of detecting a region of interest in an overall image representing a grouping of plants, wherein one or more regions of interest (201) are detected in the overall image (200) using an image processing procedure, each region of interest (201) comprising at least a part of a fruit (13),

a step (130) of acquiring local images, in which a local image is acquired in the vicinity of each region of interest (201), each local image representing at least a part of the fruit of the region of interest considered, and

a step (150) of identifying the stem or stalk, wherein a stem or stalk (14) is identified in each partial image.

2. The method according to claim 1, wherein said step (150) of identifying said stem or stalk comprises:

a sub-step (151) of detecting the elongated object, in which an element having an elongated shape is detected in each partial image, an

A discrimination sub-step in which it is determined whether each elongated object is a stem or stalk (14) or another type of elongated object.

3. The method according to claim 2, wherein the discriminating sub-step comprises:

each elongate object is classified according to its radiation intensity at a pixel or point,

the length of each of the elongated objects is determined,

determining the relative position of each elongate object with respect to the body of the fruit,

determining the curvature of each elongated object, and/or

Determining one or more angles formed between a longitudinal axis of the elongated object and a predetermined axis of the shed line.

4. The method according to any one of the preceding claims, wherein the step of acquiring a local image (130) comprises the sub-step of positioning a camera adjacent to each region of interest detected in the overall image, the camera (33A, 33B) being arranged to acquire the local image.

5. The method of any preceding claim, further comprising:

a step (140) of determining the ripeness of the fruit, wherein for the fruit of each detected region of interest (201) an indicator is determined indicating the ripeness, and/or

A step (140) of determining the health status of the fruit, wherein for the fruit of each detected region of interest (201) an indicator indicative of the health status is determined,

the method further comprises the step of selecting the fruit to be harvested, wherein the fruit to be harvested is selected in accordance with the indicator indicative of the maturity and/or the indicator indicative of the health state.

6. The method of any preceding claim, further comprising:

a step (160) of harvesting the fruit, in which the stem or stalk (14) identified in each partial image is cut, and

a step (170) of establishing a mapping of the fruits still to be harvested, wherein geographic coordinates are determined for each detected region of interest for which the fruits have not been harvested.

7. A mechanical device for harvesting fruit, the mechanical device comprising a system for selecting the fruit to be harvested, the selection system (30) comprising:

a main camera (32), the main camera (32) being arranged to acquire whole images (200), each whole image (200) representing a group of plants,

means for detecting regions of interest, the means for detecting regions of interest being arranged to detect one or more regions of interest (201) in each whole image (200), each region of interest comprising at least a part of a fruit (13),

one or more auxiliary cameras (33A, 33B), the one or more auxiliary cameras (33A, 33B) being arranged to acquire a local image in the vicinity of each region of interest (201), each local image representing at least a part of the fruit of the region of interest under consideration, and

means for identifying the stem or stalk arranged to identify the stem or stalk (14) in each partial image.

8. The mechanical device according to claim 7, further comprising one or more mechanical arms (23A, 23B), each carrying an auxiliary camera (33A, 33B) and being arranged to successively position the auxiliary cameras (33A, 33B) in the vicinity of the region of interest (201).

9. Mechanical device according to any one of claims 7 and 8, further comprising a system for harvesting fruit, said harvesting system (40) comprising at least one cutting device (41), said at least one cutting device (41) being arranged to cut the stem or stalk of the fruit.

10. Mechanical device according to claims 8 and 9, wherein a cutting means is mounted on each mechanical arm (23A, 23B) arranged to successively position the auxiliary camera carried in proximity of a region of interest and the cutting means at the stem or stalk to be cut.

11. The mechanical device according to any of the claims 7 to 10, wherein the system for selecting the fruit to be harvested further comprises a unit for determining a health status of the fruit, the unit for determining a health status of the fruit being arranged to determine an indicator indicative of the health status for each detected region of interest of the fruit.

12. The mechanical apparatus of any one of claims 7 to 11, wherein the system for harvesting fruit further comprises a gripping device mounted on each robotic arm, each gripping device being arranged to grip the stem or stalk on the fruit side before, during and after cutting.

13. The machine of claims 11 and 12, wherein the system for harvesting fruit further comprises a plurality of collection baskets and each robotic arm and the gripping device carried thereby are arranged to deposit each fruit cut by the cutting device in one of the collection baskets in accordance with an indicator determined for that fruit indicative of the health status.

14. The mechanical device of any one of claims 12 and 13, wherein the system for selecting the fruit to be harvested further comprises means for determining that an obstacle is present, the means for determining that an obstacle is present being arranged to determine whether an obstacle obstructs harvesting of the fruit.

15. The mechanical device of any one of claims 7 to 14, further comprising a purging arrangement arranged to create a gas flow in the vicinity of each region of interest.

Technical Field

The present invention relates to the field of agricultural automation, and more specifically to the field of automatically identifying the fruit to be harvested. The present invention relates to a method for harvesting fruit from stems, in particular bunches of fruit such as grape bunches, and a mechanical apparatus for harvesting fruit.

Background

Fruit harvesting, commonly referred to as "grape harvesting" for grapes, may be performed by mechanical or manual methods. The mechanical method may be performed by shaking or threshing of the grapevine seedlings, thereby separating the fruit pieces from the stems and dropping the fruit pieces onto a collection mat. This method may be well suited for certain grape varieties whose bunches do not require retention of their stalks and whose grains are not susceptible to oxidation or to the coloring of their juice by the rind. In fact, the deterioration of the berries leads to the oxidation of the sap and the dissolution of the colouring substances contained in the rind. For the above reasons, in order to avoid the fruit pieces from cracking and to preserve the stalks and their quality during the pressing, the harvesting of the grapes is carried out manually by cutting the whole cluster of fruits. This method requires considerable labor over a relatively short period of time (typically in the range of 10-15 days) during which the grapes are sufficiently ripe.

Further difficulties arise from the fact that: at harvest, vineyards have different maturity between different grapevines of the vineyard and also different maturity on the same grapevine. To address this problem, it has been suggested to integrate a fruit harvesting machine device for determining the ripeness of the fruit with a device for individually harvesting and sorting fruits that have reached a sufficient ripeness.

Application FR 2760595 a1 describes a method for harvesting grape clusters, comprising: a prior step of manually setting marking means on the stalks of the bunch that has reached a satisfactory maturity; a step of detecting the marking means and guiding the cutting device toward each marking means; and a step of cutting the carpopodium at each marking means. This solution allows harvesting only mature grapes while maintaining the integrity of the berries and the entire bunch. Nevertheless, it requires a considerable qualified labour for the setting of the marking means.

In view of the above, it is an object of the present invention to provide a fully automated method for harvesting fruit that allows the integrity of the berries and fruit clusters to be maintained. In particular, the harvesting method must allow to identify and locate the fruit to be harvested, preferably according to its maturity. Another object of the invention is to provide a process whose implementation and development costs are matched to the use in agricultural fields.

Disclosure of Invention

To this end, the invention provides a method for harvesting fruit, which comprises automatically identifying and locating the stems or stalks of the fruit, and individually and automatically cutting the fruit at these stems or stalks. The identification and localization of the stems is performed in two stages: in a first phase, a so-called overall image is acquired, representing a relatively large portion of a tree, bush or plant, and one or more fruits are detected in the overall image; in the second stage, a so-called local image is taken of each detected fruit, and the stems or stalks of these fruits are identified and positioned for their individual cutting.

More specifically, the object of the present invention is a method for harvesting fruit bearing stems or stalks, comprising:

■ detecting a region of interest in the overall image representing the grouping of plants, wherein one or more regions of interest are detected in the overall image using image processing, each region of interest comprising at least a part of a fruit,

■ acquiring local images, wherein local images are acquired in the vicinity of each region of interest, each local image representing at least a part of the fruit of the region of interest considered, and

■ identifying a stem or stalk, wherein the stem or stalk is identified in each of the partial images.

A plant group is considered to carry a group of fruits, such as a bunch of fruits. For example, a plant group is formed by a fruit tree or shrub, a climbing plant (e.g., cucumber or tomato), or a grape tree. In the latter case, the fruit to be harvested is a bunch of grapes. The overall image represents all or part of the plant grouping. As an example, the overall image may be about 1m²Area (square meters) is imaged.

The overall image may be a two-dimensional or three-dimensional image. Typically, a two-dimensional image is formed using a camera comprising a sensor formed by a set of photosensitive elements (or pixels) organized in a matrix. Each pixel may be associated with an intensity of radiation in multiple bands of the visible, ultraviolet, and/or infrared spectrum. The visible spectrum is defined as the electromagnetic spectrum with wavelengths comprised between 390nm (nanometers) and 780 nm; the ultraviolet spectrum is defined as the electromagnetic spectrum with wavelengths comprised between 10nm and 390 nm; and the infrared spectrum is defined as the electromagnetic spectrum with wavelengths comprised between 780nm and 0.1mm (millimeters). In particular, each pixel may be associated with a radiation intensity in a band of the near infrared spectrum (comprised between 780nm and 3 μm (micrometer)). For example, the overall image is a multispectral image, that is to say data captured in a plurality of generally separate wavelength bands, or a hyperspectral image, that is to say data captured in a plurality of wavelength bands covering a wide wavelength range.

The three-dimensional image comprises a point cloud, each point being defined by a location in space. Each point of the three-dimensional image may also be associated with an intensity of radiation in multiple bands of the visible, ultraviolet, and/or infrared spectrum.

Image processing for detecting regions of interest in the overall image may include supervised or unsupervised classification methods based on shape and color.

A partial image is acquired to enable the stem or stalk of the fruit to be identified and located. Thus, the local image is acquired with a narrow field of view, indicating only one fruit, i.e. the detected fruit of the region of interest. Still by way of example, each partial image may be 300cm²The area of the range is imaged. In this sense, the stem of the fruit is generally located above the main part of the fruit by the action of gravity, preferably the local image is taken in a downward view, that is, below the horizontal line, so that the stem in the local image is visible.

Each partial image may be a two-dimensional image or a three-dimensional image. Each two-dimensional partial image comprises a set of pixels, each pixel being defined by radiation intensities in a plurality of bands of the visible spectrum and/or the infrared spectrum. Each three-dimensional partial image comprises a point cloud, each point being defined by a position in space and by radiation intensities in a plurality of bands of the visible and/or infrared spectrum.

The step of identifying the stem or stalk may comprise a sub-step of detecting the elongated object, wherein an element having an elongated shape is detected in each partial image, and a sub-step of discriminating, wherein it is determined whether each elongated object is a stem or stalk or another type of elongated object. An element is defined as having an elongated shape when the ratio of its largest dimension to its orthogonal dimension or each of two other orthogonal dimensions is greater than or equal to 3.

The discriminating sub-step may be performed using different criteria. These criteria may cover the electromagnetic spectrum reflected by the elongated objects, the dimensions of these elongated objects, their shape and their orientation. These criteria may be used alone or in combination.

According to a first variant, the recognition sub-step comprises classifying each elongated object according to the radiation intensity of the pixels or points of the elongated object. In particular, during said classification sub-step, it can be determined whether the spectrum of each elongated object is at least partially located in a band corresponding to green or brown. In the case of grape bunches, green generally corresponds to the fruit stalks and brown to the grape branches. During the sub-step of classifying the elongated objects, it may also be determined whether the spectrum of each elongated object is at least partially located in a near infrared band showing the presence or absence of chlorophyll, for example, a band comprised between 680nm and 850 nm.

According to a second variant, the recognition sub-step comprises determining the length of each elongated object. An elongated object may be considered a stem or a stalk if its length is comprised in a predetermined length range. For a bunch of grapes, the range is comprised, for example, between 0.5cm (centimetres) and 3 cm. Lengths longer than 3cm typically correspond to trellis lines.

According to a third variant, the distinguishing sub-step comprises determining the relative position of each elongated object with respect to the body of the fruit. Thus, if an elongated object is located above the body of the fruit, the elongated object may be considered to be a stem or stalk.

According to a fourth variant, the recognition sub-step comprises determining the curvature of each elongated object. The likelihood of the stem or stalk extending vertically downwards is relatively low and the cluster of fruit often causes the stem or stalk to bend due to its own weight. Thus, when an elongated object has a curvature greater than a predetermined threshold, the elongated object may be considered a stem or a stalk.

According to a fifth variant, the discrimination sub-step comprises determining one or more angles formed between the longitudinal axis of each elongated object and the predetermined axis of the shed line. The plants may be aligned to form rows aligned parallel to each other. In each row, one or more canopy lines may be installed to serve as a support frame for the plants. Each trellis line then forms an elongated object that is likely to be identified as a fruit stem or stalk. In order to exclude these shelving lines from the detected elongated objects, each elongated object whose longitudinal axis forms at least one angle smaller than the predetermined angle may be considered a shelving line. In contrast, when each or at least one angle formed between the longitudinal axis of the elongated object and the predetermined axis of the shed line is larger than a predetermined angle, for example equal to 10 degrees, the elongated object may be considered to be a stem or a stalk.

The step of acquiring a partial image may comprise the sub-step of positioning a camera adjacent to each region of interest detected in the overall image, the camera being arranged to acquire the partial image. Preferably, a camera is positioned above each fruit so that the stem or stalk bearing the fruit is visible.

According to a particular embodiment, the method further comprises a step of determining the ripeness of the fruit, wherein an indicator indicative of the ripeness index is determined for each detected fruit of the region of interest. In particular, an indicator indicative of the ripeness of the fruit may be determined from the radiation of the fruit in one or more predetermined wavelength ranges. These wavelength ranges may show sugar content and/or anthocyanin content.

Still according to a particular embodiment compatible with the previous embodiment, the method further comprises a step of determining the health status of the fruit, wherein for each detected fruit of the area of interest an indicator indicative of the health status is determined. In particular, the indicator indicative of the health status may be indicative of the presence of a pest in or on the fruit. For example, the pest is a plant pathogen (bacteria, viruses, fungi, etc.) or a pest (molluscs, arachnids, invading insects, etc.). The indicator indicative of the health status may be determined from the radiation of the fruit in one or more predetermined wavelength ranges. These wavelength ranges may indicate the presence of harmful substances, for example, powdery mildew and/or mold and/or botrytis. The health status indicator may also be determined by shape recognition. In particular, for a grape bunch, the health indicator may be determined by identifying the shape of the berries of the grape bunch.

An indicator indicative of maturity and/or an indicator indicative of health status may be determined for each fruit in the global image and/or the local image.

The method may also comprise the step of selecting the fruit to be harvested, wherein the fruit to be harvested is selected according to an indicator indicating maturity and/or an indicator indicating health status.

The method may also include the step of determining the presence of an obstacle, wherein it is determined whether the obstacle obstructs harvesting of the fruit. The step of determining the presence of an obstacle may be performed for each fruit in the overall image or, preferably, for each fruit in the corresponding partial image. During the step of selecting the fruit to be harvested, the fruit to be harvested may then also be selected according to the presence or absence of obstacles.

Advantageously, the method further comprises a step of harvesting the fruit, wherein the stem or stalk identified in each partial image is cut. Preferably, each stem or stalk is cut immediately after it is identified. In case the method comprises a step of selecting the fruit to be harvested, only the selected fruit is harvested during the step of harvesting the fruit.

The method may also comprise the step of establishing a mapping of the fruits still to be harvested, wherein geographic coordinates are determined for each detected region of interest for which a fruit has not been harvested. This is the case, for example, when the stem or stalk of the fruit cannot be identified, when the maturity index of the fruit is not appropriate, or when an obstacle prevents harvest. The mapping of the fruits still to be harvested allows to determine the number and the position of the fruits that need to be harvested manually and thus allows to plan the required manual work.

The method may also comprise the step of establishing a fruit map, wherein geographical coordinates are determined for each detected region of interest. This mapping allows the distribution and density of the fruit over an area to be determined.

The object of the invention is also a mechanical device for harvesting fruit, which is arranged to carry out the above-mentioned method. In particular, the mechanical device may comprise a system for selecting fruits to be harvested, the system comprising:

■ a main camera arranged to acquire whole images, each whole image representing a grouping of plants,

■ for detecting regions of interest, the unit being arranged to detect one or more regions of interest in each whole image, each region of interest comprising at least a part of the fruit,

■ one or more auxiliary cameras arranged to acquire local images in the vicinity of each region of interest, each local image representing at least part of the fruit of the region of interest under consideration, and

■ for identifying a stem or stalk, the unit being arranged to identify the stem or stalk in each of the partial images.

The mechanical device may be provided with propulsion means to ensure its movement. The mechanical device forms an autonomous robotic machine. The mechanical equipment may also be mounted on an agricultural tractor or a running support frame intended to be towed.

Each (primary and secondary) camera may comprise a sensor formed by a set of photosensitive elements (or pixels) organized in a matrix. Preferably, each camera is arranged to acquire images over at least a portion of the visible spectrum. The camera may also be arranged to further cover a part of the ultraviolet and/or infrared spectrum.

Each auxiliary camera may also be a stereoscopic camera, that is to say comprising two sensors arranged so as to be able to construct three-dimensional images, each comprising a point cloud, each point being defined by a position in space. Alternatively, each auxiliary camera may be a plenoptic camera.

The robotic device may further comprise one or more robotic arms, each carrying an auxiliary camera and arranged to successively position the auxiliary camera in the vicinity of the region of interest. For example, the mechanical apparatus includes three or four mechanical arms. The robotic arm is controlled such that for each region of interest, one auxiliary camera acquires a local image.

According to a particular embodiment, the mechanical device further comprises a system for harvesting the fruit. The harvesting system comprises at least one cutting device arranged to cut the stem or stalk of the fruit.

According to a first variant, a cutting device is arranged on each mechanical arm. Each robotic arm is then arranged to successively position the carried auxiliary camera in the vicinity of the region of interest and to position the cutting device at the stem or stalk to be cut.

According to a second variant, each cutting device is mounted on a dedicated cutting robot arm arranged to successively position the cutting device at each identified stem or stalk, so as to effect the cutting thereof.

According to a particular embodiment, the system for selecting a fruit to be harvested further comprises:

■ for determining the ripeness of the fruit, the unit being arranged to determine the ripeness for each detected fruit of the region of interest, and/or

■ for determining the health status of the fruit, the unit being arranged to determine a health status indicator for each detected fruit of the region of interest.

The system for harvesting fruit may further comprise a gripping device mounted on each robotic arm, each gripping device being arranged to grip the stem or stalk on the fruit side before, during or after cutting. The system for harvesting fruit may also include, where appropriate, a gripping device mounted on each cutting robot arm. Advantageously, the gripping means are mounted on the same arm as the cutting means.

The system for harvesting fruit may also include one or more baskets for collecting cut fruit.

According to a first variant, the system for harvesting the fruits comprises a single collecting basket, and each mechanical arm and the gripping device carried are arranged to deposit each fruit cut by the cutting device in the collecting basket. For example, the collecting basket is fixed to the frame of the mechanical device.

According to a second variant, the system for harvesting fruit comprises a plurality of collecting baskets and each mechanical arm and the gripping device carried are arranged to deposit each fruit cut by the cutting device in one collecting basket according to a determined health indicator of the fruit. This variant allows to automatically sort out the intact fruits from the infected or damaged fruits. For example, the collecting basket is fixed to the frame of the mechanical device.

According to a third variant, the system for harvesting fruit comprises a system mounted on each mechanical arm carrying a cutting device, below said cutting device.

The system for selecting fruit to be harvested further comprises a unit for determining the presence of an obstacle, the unit being arranged to determine whether the obstacle obstructs harvesting of the fruit. In particular, the means for determining the presence of an obstacle may be arranged to determine whether there is sufficient space between the branches of the plant and the trellis line (where appropriate) to enable the robotic arms and cut fruit to pass through. For example, the presence of an obstacle is determined from the whole image or the partial image.

Furthermore, the mechanical apparatus may further comprise a purging device arranged to form a gas flow in the vicinity of each region of interest. Preferably, an air flow is created near each region of interest when the local image is acquired near the region of interest, and/or when the cutting operation is performed. For example, the purge means includes a central purge mounted on the frame of the mechanical apparatus, a blow nozzle mounted on each mechanical arm, and an air conduit connecting each blow nozzle to the central purge. In other embodiments, the purge device may include a purge machine mounted on each robotic arm.

According to a particular embodiment, the mechanical device is arranged to be able to harvest the fruit during its movement. In particular, the robotic arm may be arranged to compensate for the overall movement of the mechanical device during the operations of acquiring partial images, holding and cutting the stem or stalk of the fruit.

Drawings

Other features, details and advantages of the invention will appear, upon reading the following description, which is provided by way of example only and with reference to the accompanying drawings, in which:

fig. 1A schematically shows in perspective an example of a mechanical device for harvesting a bunch of grapes in a vineyard;

FIG. 1B schematically shows the mechanical apparatus of FIG. 1A in a vineyard in side view;

fig. 2 shows an example of a method for harvesting fruit according to the present invention;

FIG. 3 shows the result of the step of detecting a region of interest in the method of FIG. 2;

fig. 4 shows an example of the implementation of the step of identifying the stems in the method of fig. 2.

Detailed Description

Fig. 1A and 1B schematically show an example of a mechanical device for harvesting fruit according to the present invention. In FIG. 1A, the mechanical device is shown in perspective view, and in FIG. 1B, the device is shown in side view in a row according to an axisIn extended grapevine. In the examples described below, mechanical equipment is used to harvest the bunch of grapes in the vineyard. Nevertheless, the invention is also applicable to the harvesting of any other fruit. The grapevine 10 shown in fig. 1B includes a trunk 11 from which grape branches 12 extend and fruit clusters 13 each connected to a grape branch 12 by a fruit handle 14. Mechanical equipment 20 quiltArranged to identify the stem 14 of each cluster 13 and cut it. The device comprises a frame 21 mounted on an axle 22 to enable movement of said frame 21. Machine 20 may be towed manually or by a tractor, or equipped with propulsion devices, to ensure autonomous movement. Mechanical apparatus 20 further comprises: two robot arms 23A, 23B; a system 30 for selecting fruits to be harvested; a system 40 for harvesting fruit; and a control unit 50.

Each robotic arm 23A, 23B is mounted on the frame 21 and includes a distal end adapted to move relative to the frame 21 according to a plurality of degrees of freedom. In the embodiment of fig. 1A and 1B, each robot arm 23A, 23B comprises according to an axisExtended first section 231, according to axisA second section 232 pivotably connected to the first section 231 and according to a planeIs pivotally connected to the third section 233 of the second section. Each robot arm 23A, 23B further includes: the first segment 231 is carried so as to be able to follow the axisA first guidance system 234 for translational movement; the first guide system 234 is carried so as to be able to follow the axisA second guidance system 235 for translational movement; and a second guide system 235 so as to enable it to follow the axisA third guide system 236 that moves in translation. Thus, the third section 233 of each robot arm 23A, 23B is adapted to move relative to the frame 21 according to five degrees of freedom. Thus, each robot arm 23A, 23B may be referred to as "5-axis mechanical arm ". Of course, the mechanical apparatus for harvesting fruit according to the present invention may comprise any type of mechanical arm, in particular a mechanical arm with a greater number of degrees of freedom. The third section 233 of each robot arm 23A, 23B is arranged to carry a camera and a tool as explained below.

The selection system 30 includes a main camera 32 and auxiliary cameras 33A, 33B for each robot arm 23A, 23B. The main camera 32 is arranged to acquire a so-called overall image of one or more grapevines 10 or of at least a part of one or more grapevines 10. The main camera 32 may be fixedly mounted on the frame 21. For example, it is oriented such that its line of sight is aligned with the axisAre orthogonal. The main camera 32 is a so-called RGB camera, that is to say, comprises a set of photosensitive elements adapted to distribute the light intensity in at least three bands of the visible spectrum to each point of the image space. The master camera 32 may also cover a portion of the ultraviolet and/or infrared spectrum. Each robot arm 23A, 23B carries an auxiliary camera 33A, 33B, respectively, on its segment 233. As explained below, it is arranged so that the auxiliary cameras 33A, 33B can be positioned at desired successive positions of the vineyard, respectively. Each auxiliary camera 33A, 33B is arranged to acquire a so-called partial image, typically covering a single fruit string or fruit string portion. For example, the auxiliary cameras 33A, 33B are constituted by stereo cameras, that is to say each is equipped with two photosensitive sensors arranged to acquire two images of the same scene according to different angles and thus generate a three-dimensional (3D) image. Each photosensitive sensor may be arranged to acquire radiation intensity in a plurality of wavelength bands in the visible spectrum, ultraviolet spectrum and/or infrared spectrum. Thus, the stereo camera allows for simultaneous collection of spectral and positional information in the observed scene. In particular, each 3D partial image may consist of a point cloud, each point being associated with a radiation intensity in a plurality of wave bands.

For each robot 23A, 23B, the harvesting system 40 comprises a cutting device 41 and a gripping device 42 mounted on the segment 233 of the robot in question. The recovery system 40 also includes two collection baskets 43A, 43B. The cutting means 41 is arranged to cut the fruit stalks 14 and the gripping means 42 is arranged to grip each stalk when cutting. Collecting baskets 43A, 43B are mounted on frame 21 and each is arranged to receive a string of cut fruit.

The control unit 50 is arranged to receive the overall image generated by the main camera 32 and the local images generated by each of the auxiliary cameras 33A, 33B. The control unit 50 is also arranged to process these images and control the robot arms 23A, 23B, the cutting device 41 and the gripping device 42, as explained below.

The mechanical device according to the invention may also comprise a geographical positioning system and a mapping unit, not shown. The geographical positioning system is arranged to determine geographical coordinates of the mechanical device or one of its components. In particular, the mechanical device may comprise a geographic positioning system mounted on each mechanical arm 23A, 23B, in order to determine the geographic coordinates of each auxiliary camera 33A, 33B and, therefore, of each object imaged by these cameras. Each geolocation system may include a satellite positioning system and/or an inertial unit. The mapping unit is arranged to generate a map defining the positions of the detected fruits in the overall image. As explained below, the mapping may take into account different criteria, such as the maturity and/or health status of the fruit, and/or the fact that the stem of the fruit cannot be identified, where appropriate.

Fig. 2 shows an example of a method for harvesting fruit according to the invention, and fig. 3 shows the result of one step of the method, wherein a region of interest is detected. The method is described with reference to a mechanical device (described with reference to fig. 1). The method 100 comprises: a step 110 of acquiring a whole image; a step 120 of detecting a region of interest; a step 130 of acquiring a partial image; a step 140 of determining a maturity index and a health status index of the fruit; a step 150 of identifying stems; a step 160 of cutting the stems; a step 170 of storing the cut fruit bunch; and a step 180 of establishing a mapping of the fruit. During the step 110 of acquiring a whole image, the main camera 32 acquires a whole image, which has the reference number 200 in fig. 3, and transmits it to the control unit 50. During step 120, the control unit 50 performs image processing, including detecting regions of interest in the overall image 200, each region of interest comprising a bunch of grapes. In fig. 3, the region of interest has reference numeral 201. The step 120 of detecting the region of interest may comprise a classification method based on the shape and color of the fruit cluster to be harvested. As shown in fig. 3, the fruit string may only be partially shown in the overall image, and the region of interest 201 may only include a visible portion of the fruit string 13. Then, steps 130, 140, 150, 160, 170 are performed for each region of interest detected in the overall image. During step 130, by actuating the corresponding robot arm 23A, 23B, one of the auxiliary cameras 33A, 33B is positioned in the vicinity of the region of interest 201 under consideration, and a 3D local image is acquired for this region of interest and transmitted to the control unit 50. During step 140, for each detected region of interest, the control unit 50 determines a maturity index and a health status index of the fruit cluster. These indicators are determined from the intensity of radiation in one or more predetermined wavelength ranges of the corresponding 3D partial images. When the maturity is deemed satisfactory for the region of interest under consideration, the method continues with step 150 for that region of interest. Conversely, when the index is not satisfactory for the region of interest considered, then said region of interest is excluded for the remaining steps of the method. Step 150 includes identifying the stalk of each fruit string using the corresponding 3D partial image. During step 160, the robotic arms 23A, 23B for positioning the auxiliary cameras (which have acquired the 3D partial images) position the cutting device 41 and the gripping device 42 at each identified fruit handle; actuating the gripping means 42 to facilitate gripping of the fruit stem; the cutting device 41 is then activated so as to cut the fruit stem upstream of the gripping device 42. The step 170 of storing the cut fruit bunch includes storing the cut fruit bunch in one of the collection baskets 43A, 43B according to its health indicator. The step 170 is performed by actuating the corresponding robotized arm 23A, 23B, the gripping device 42 performs the gripping of the fruit stalks until the cut cluster reaches the selected collection basket 43A, 43B. The step 180 of establishing a mapping of the fruit comprises determining the geographical coordinates of all fruits detected in the overall image or parts thereof which fulfill a predetermined condition. As an example, a mapping may be established for fruits characterized by insufficient maturity or fruits for which the stem cannot be identified during step 150. The method 100 may then be repeatedly restarted while a new overall image representing another portion of the vineyard is acquired.

It should be noted that step 110-170 in the method for harvesting fruit according to the present invention may be performed while mechanical apparatus 20 is stopped or moving (e.g., moving at a relatively constant speed). In case the method is performed during movement of the mechanical device, the robotic arm may then be arranged to compensate for the movement of the mechanical device, in particular for the step 130 of acquiring the partial image and the step 160 of cutting the stem.

Fig. 4 shows an example of carrying out the step 150 of identifying stems. In this example, step 150 comprises a sub-step 151 of detecting the elongated object and a sub-step 152 of classifying the elongated object. Sub-step 151 comprises detecting elements having an elongated shape in each 3D partial image. For example, an element is defined as having an elongated shape when the ratio of its largest dimension to each of the other two orthogonal dimensions is greater than or equal to a predetermined threshold. For example, the threshold is set equal to 3. Sub-step 152 includes determining whether each elongated object is a stalk or other type of object. For example, substep 152 may be performed by checking that the elongated object is free of branches. Substep 152 may also be performed by determining whether the spectrum of each elongated object is at least partially within a predetermined band of wavelengths. In particular, for a grape bunch, green elongate objects may be classified as a stalk class, brown elongate objects may be classified as a grape branch class, and gray elongate objects may be classified as a trellis line class. The sub-step 152 of classifying the elongated objects may also be performed by determining the length, curvature and/or orientation of each elongated object.

It should be noted that the control unit 50 performs a first image processing including detecting a region of interest in the whole image and a second image processing including identifying a stem (or fruit stalk) in the partial image. The control unit 50 also determines the maturity and health indicators of the fruit and the presence of an obstacle. Thus, the control unit 50 can be considered to comprise a unit for detecting a region of interest, a unit for identifying the stem, a unit for determining the maturity of the fruit, a unit for determining the health status of the fruit and a unit for determining the presence of an obstacle. Further, the control unit 50 performs control of the robot arms 23A, 23B, the cutting device 41, and the gripping device 42. The control unit 50 also compensates, where appropriate, for the movements of the mechanical device during the identification and cutting of the stem. In practice, the control unit 50 may thus be formed by a plurality of different hardware and/or software elements.