阅读说明：本技术 用于自动化车辆的对象检测系统 (Object detection system for an automated vehicle ) 是由 伊扎特 袁萍 于 2017-04-17 设计创作，主要内容包括：本发明提供一种用于自动化车辆的对象检测系统(10),所述对象检测系统(10)包括对象检测器(20)、数字地图(14)和控制器(12)。所述对象检测器(20)用于观察主车辆(22)附近的视野(32)。所述数字地图(14)用于指示所述主车辆(22)附近的道路特征(56)。所述控制器(12)被配置来基于所述道路特征(56)而将关注区域(24)限定在所述视野(32)内,并且优先处理来自所述对象检测器(20)的对应于所述关注区域(24)的信息。(An object detection system (10) for an automated vehicle is provided, the object detection system (10) including an object detector (20), a digital map (14), and a controller (12). The object detector (20) is for observing a field of view (32) in the vicinity of a host vehicle (22). The digital map (14) is used to indicate road features (56) in the vicinity of the host vehicle (22). The controller (12) is configured to define a region of interest (24) within the field of view (32) based on the road feature (56), and to prioritize information from the object detector (20) corresponding to the region of interest (24).)

1. A system (10) of a host vehicle (22), the system comprising a controller (12), the controller (12) configured to:

maintaining a digital map (14) indicative of road features (56) in the vicinity of the host vehicle (22);

defining a region of interest (24) within a field of view (32) proximate the host vehicle based on the road feature (56);

selecting a sensor (36) type for the area of interest (24) within the field of view (32) from a plurality of sensor types (36) based on the road feature; and

preferentially processing information obtained from the selected type of sensor (36) corresponding to the area of interest (24) over processing other information collected from other sensor types of the plurality of sensor types.

2. The system of claim 1, wherein the plurality of sensor types includes lidar, camera, and radar.

3. The system of claim 1, wherein the controller (12) is further configured to define the area of interest (24) based on a speed of the host vehicle.

4. The system of claim 1, wherein the controller (12) is further configured to change an update rate of the selected type of sensor (36) based on the road characteristic (56).

5. The system of claim 1, wherein the controller (12) is further configured to vary an angular resolution of the selected type of sensor (36) based on the road characteristic (56).

6. The system of claim 1, wherein the controller is further configured to extend a range of the selected type of sensor (36) based on the road feature (56).

7. The system of claim 1, wherein the controller is further configured to increase a signal-to-noise ratio of the selected type of sensor (36) based on the road characteristic (56).

8. The system of claim 1, wherein the controller is further configured to change a camera angle of the selected type of sensor (36) based on the road feature (56).

9. The system of claim 1, wherein the controller is further configured for selecting a type of lane marking detection algorithm to be used for running on the information based on the road characteristic (56).

10. A method, comprising:

maintaining, by a controller of a system of a host vehicle, a digital map (14) indicative of road features (56) in proximity to the host vehicle (22);

defining a region of interest (24) within a field of view (32) proximate the host vehicle based on the road feature (56);

selecting a sensor (36) type for the region of interest (24) within the field of view (32) from a plurality of sensor types (36) based on the road feature (56); and

preferentially processing information obtained from the selected type of sensor (36) corresponding to the area of interest (24) over processing other information collected from other sensor types of the plurality of sensor types.

11. The method of claim 10, wherein the controller comprises the controller of any one of claims 1-9.

12. A computer readable storage medium comprising instructions that, when executed, configure a controller of a system of a host vehicle to perform the method of claim 10 or claim 11.

13. A system comprising:

means for maintaining a digital map (14) indicative of road features (56) in the vicinity of the host vehicle (22);

means for defining a region of interest (24) within a field of view (32) in proximity to the host vehicle based on the road feature (56);

means for selecting a sensor (36) type for the region of interest (24) within the field of view (32) from a plurality of sensor types (36) based on the road feature (56); and

means for prioritizing processing of information obtained from the selected type of sensor (36) corresponding to the area of interest (24) over processing of other information collected from other sensor types of the plurality of sensor types.

Technical Field

The present disclosure relates generally to an object detection system for an automated vehicle, and more particularly to a system that defines a region of interest within a field of view of an object detector based on road characteristics and prioritizes information from the region of interest.

Background

It is known to equip automated vehicles with sensors to observe or detect objects in the vicinity of the automated vehicle. However, the processing power required to process all the information available from the sensors about the entire area around the automated vehicle makes the cost of the processing equipment prohibitively high.

Summary of The Invention

According to one embodiment, an object detection system for an automated vehicle is provided. The system includes an object detector, a digital map, and a controller. The object detector is for observing a field of view in the vicinity of the host vehicle. Digital maps are used to indicate road features in the vicinity of the host vehicle. The controller is configured to define a region of interest within the field of view based on the road feature and to prioritize information from the object detector corresponding to the region of interest.

Further features and advantages will appear more clearly on reading the following detailed description of preferred embodiments, given purely by way of non-limiting example and with reference to the accompanying drawings.

Brief Description of Drawings

The invention will now be described, by way of example, with reference to the accompanying drawings, in which:

FIG. 1 is a block diagram of the system of the present invention;

FIG. 2 is a schematic diagram depicting sensor coverage implemented by the system of FIG. 1 in the vicinity of a host vehicle;

FIG. 3 is a schematic diagram depicting an example implementation of the system of FIG. 1 using a centralized controller;

FIGS. 4A and 4B are schematic diagrams depicting the adjustment of angular resolution in a ROI by the system of FIG. 1; where fig. 4A is a plot of one quarter of the FOV of the sensor and the update rate of the sensor is reduced by one third, and fig. 4B is a plot of FOV/image resolution. Fig. 5A and 5B depict schematic diagrams of the signal-to-noise improvement achieved by the system of fig. 1 by averaging, where measured data prior to averaging is shown in fig. 5A and data after averaging is shown in fig. 5B.

Detailed Description

FIG. 1 shows a non-limiting example of an object detection system 10 (hereinafter system 10). The system 10 is suitable for use on an automated vehicle (hereinafter host vehicle 22). The system 10 includes an object detector 20, which object detector 20 may include various sensors 36, which sensors 36 are used to view the field of view 32 for detecting objects in the vicinity of the host vehicle 22. By way of example, but not limitation, the sensors 36 in the object detector 20 may include a camera, a radar unit, a lidar unit, or any combination thereof. The controller 12 may also include or be in communication with a vehicle sensor 16, the vehicle sensor 16 being adapted to measure a velocity 50 of the host vehicle 22 and a yaw rate 52 of the host vehicle 22. Information from the sensors 36 of the object detector 20 may be processed by the object test 18 in the controller 12 to detect objects 58 in the field of view 32.

The system 10 also includes a digital map 14, the digital map 14 indicating road features 56 in the vicinity of the host vehicle 22. The digital map 14 and the vehicle sensors 16 are used to define the type of environment and mode surrounding the host vehicle 22. The host vehicle 22 is positioned on the digital map 14 using the map positioning 62 in the controller 12.

The controller 12 is configured to define the region of interest 24 within the field of view 32 based on the road features 56 and to prioritize information from the object detector 20 corresponding to the region of interest 24. As used herein, the road features 56 may define a subset of the digital map 14 that includes lane and road attributes; and preferential processing may indicate focusing on the region of interest 24 to facilitate acquisition of denser and more accurate sensor data, processing data within the region of interest 24 at a higher rate, assigning more processing and communication resources to the region of interest 24, and adjusting parameters and algorithms of the subject test 18 within the region of interest 24.

Advanced Driver Assistance Systems (ADAS) and automated vehicles are equipped with various sensors 36, such as lidar units, radar units, and/or cameras to view the area around the host vehicle 22. The field of view (FOV)32 of these sensors 36 may cover up to 360 ° of the area around the host vehicle 22. These sensors 36 are used to detect objects 58 around the host-vehicle 22 and to make decisions about actions to take based on the environment around the host-vehicle 22. The use of these sensors 36 places a significant burden on the processing and communication resources of the host vehicle 22 because of the large amount of data that needs to be captured by the sensors 36, the data that needs to be transmitted to a processing unit, and the data that needs to be processed by an onboard processing unit of the host vehicle 22 to enable object 58 detection and other functions. This therefore increases the complexity and cost of the system 10. A method is presented for selecting a region of interest 24 (hereinafter ROI 24) for centralized processing based on road features 56 as determined by a digital map 14 device. For example, if on a highway, processing may be focused on the front of the host vehicle 22, more processing and communication resources may be allocated to the data stream from the front sensors.

To overcome the shortcomings of the sensors 36, the digital map 14 currently plays an important role in many ADAS and autonomous vehicle systems. The digital map 14 provides valuable information that may be used for control and path planning, among other applications. The information provided by the digital map 14 varies from map provider to map provider. In automotive applications, the digital map 14 provides geometric information and other attributes about the road. In general, the output from the digital map 14 may include, but is not limited to: a map describing future points of the road, the curvature of the road, the lane marker type, the lane width, the speed 50 limit, the number of lanes, the presence of exit ramps, obstacles, marker locations, etc. The digital map 14 may be used for various tasks, such as improving the perception algorithm by using the digital map 14 as a priori information or by treating the digital map 14 as a virtual sensor. A subset of digital map 14 information, including geometric and road attributes, in the vicinity of the host vehicle 22 is used to define an ROI 24 around the host vehicle 22, the digital map 14 information defining the environment around the host vehicle 22. This subset of information will be referred to as road characteristics 56. It should be noted that the road features 56 are only a subset of the digital map 14, as some digital map 14 information, such as highway names and numbers, are not needed for the purpose of defining the ROI 24. ROI 24 concentrates sensor acquisition and processing on a small area and thus provides significant savings in processing and communication requirements.

Fig. 2 shows an example of the host vehicle 22 equipped with a 360 ° field of view 32. The figure shows an example of ROI 24 selected on a highway. The figure also shows an example of a curved road 30 and an example of an intersection 26, in which case the processing should focus on certain angles relative to one side of the host vehicle 22. It is important that the regions outside ROI 24 are not completely ignored, as they may contain important information about host-vehicle 22. Depending on the processing and communication capabilities of the host vehicle 22, other sectors of processing may be prioritized for lower rates.

There are a variety of methods that may be used to define ROI 24 based on road features 56. In one embodiment, the sensors 36 of the object detector 20 are collected in a distributed manner by a plurality of devices prior to communicating with the controller 12. The road features 56 may be delivered to the host vehicle 22 using an external link or stored in the host vehicle 22 according to a previously defined path. In a preferred embodiment of the present invention, the sensors 36 of the object detector 20 are collected using a centralized method 40 as shown in FIG. 3. In fig. 3, the output of the sensor 36 is directed into the controller 12 using ethernet or other connector types. Controller 12 then decides which portions of sensors to retain and which portions of sensors to discard based on the selected region of ROI 24.

In another possible variation, the controller 12 sends a signal to the sensor 36 to turn the sensor on and off as needed. An advantage of this approach is that it may save power, but may not be feasible for many sensors 36. Based on knowledge of sensors 36 in subject detector 20, controller 12 may choose to combine the two approaches described above, with the electrically controllable sensors 36 outside of ROI 24 being turned off, while for other sensors 36, controller 12 ignores or maintains sensor measurements as required by the definition of ROI 24.

The speed 50 of host-vehicle 22 has a significant impact on ROI 24 for proper object 58 detection. The 3 second rule is widely used in automobile following. The rules are typically used to check the amount of space left on the front side of the host-vehicle 22 so that the driver can be prepared for braking if the vehicle on the front side of the host-vehicle is stopped or slowed. The 3 second rule can be significantly affected by road conditions and visibility. As an example, the 3 second rule may double in rainy, foggy, snowy, night, etc. days. In one embodiment, range 34 of ROI 24 is determined via speed 50 of host-vehicle 22 using a 3-second rule as a criterion. In this method, the velocity 50 of the host vehicle 22 is used to determine the range 34 of the ROI 24 using the formula 3 × m/sec, where 3 is calculated from the 3 second rule and the velocity 50 of the meter/sec autonomous vehicle 22. As one example, for a host vehicle 22 traveling at a speed 50 of one hundred kilometers per hour (100kph), the range 34 of ROI 24 should be around eighty-five meters (85 m). The extent 34 of ROI 24 may be smaller at lower velocities 50. FIG. 2 shows an example of high velocity ROI 24 and low velocity ROI 28. The range 34 in high-speed ROI 24 may extend up to the maximum range of the sensor. The FOV of the lower-speed ROI 28 may be increased, if necessary. It should be noted that it is straightforward to extend range 34 of ROI 24 using weather information such as rain (as an example). Using the example above, ROI 24 would extend to 170 meters in the case of rain. Rain sensing may be accomplished using rain sensors of the host vehicle 22 that are widely used to control wipers.

Another factor affecting ROI 24 is the yaw rate 52 of host vehicle 22. Most instances of the host vehicle 22 are equipped with sensors to measure the angular velocity of the host vehicle about its vertical axis (referred to as the yaw rate 52). Controller 12 should use yaw rate 52 to determine the direction of ROI 24 near host-vehicle 22. When the host-vehicle 22 turns left or right, the ROI 24 should be adjusted to align with the host-vehicle 22. Fig. 2 shows an example where ROI 24 is centered on the right side 30 of host vehicle 22. The sensors used in ROI 24 may be selected from sensors 36 in ROI 24 or rotated to better match ROI 24. Similarly, ROI 24 may be adjusted based on the curvature of the road as determined from road feature 56.

In a typical object 58 detection system 10, multiple classes of objects are detected, such as vehicles, pedestrians, and bicycles. The number of classes of objects 58 may become very large, which places many demands on the processing and communication requirements of the host vehicle 22. Limiting the number of types of objects 58 to detect may result in significant savings in processing and communication requirements for the host-vehicle 22.

In one embodiment, road feature 56 provides controller 12 attributes to help decide what to run object test 18 in ROI 24 and how often object test 18 is run in ROI 24. As an example, one of the attributes of the road feature 56 is the type of lane marking in the vicinity of the host vehicle 22. There are many types of lane markings, such as guide points, solid lines, or dashed lines. The algorithms used to detect these types of lane markers may vary significantly. Thus, the controller 12 may access such information from the road features 56 and make a decision on the type of lane marking detection algorithm to run in the subject test 18. In addition to the lane marking algorithm type, the road features 56 may also provide information to adjust parameters of the algorithm based on map information. As an example, based on the road type, the width of the lane may be determined, which may vary between highways, residential roads, etc.

In another embodiment, the number of algorithms to be run may be adjusted based on attributes from the road characteristics 56. As an example, if the map indicates that the host vehicle 22 is currently on a restricted access highway, then the likelihood of a pedestrian or bicycle being present is low. Thus, the pedestrian or bicycle detection algorithm is not run or executed at a reduced rate. This may result in substantial savings in the processing requirements of the host vehicle 22.

In addition to processing power savings and sensor selection, ROI 24 selection may also be used to enhance sensor output. As an example, there is a trade-off between FOV/image resolution 48 versus range 34 of the sensor. In the case of incorporating map information, the trade-off may change dynamically. For example, the FOV of the sensor may increase as the image resolution 48 increases, while decreasing the range 34 in urban areas (or vice versa on public roads). This can bring significant benefits to processing and algorithm performance.

ROI 24 may be dynamically assigned a higher angular resolution 48 and update rate 54 while maintaining a lower angular resolution 48 and update rate 54 for monitoring in regions outside of ROI 24. For example, the FOV may increase as the image resolution 48 increases, while decreasing the range 34 in urban areas; on the other hand, in road driving, the range 34 may be increased and the FOV may be decreased and made to follow a path according to a map and/or an object acquired by a monitoring function. ROI 24 may be implemented by reducing and enlarging optics in object detector 20 during operation and/or dynamically changing a scanning mode of a scanning lidar in object detector 20.

In one embodiment, the sensor update rate 54 is controlled based on road characteristics. The main idea is based on the following facts: ROI 24 is important and therefore requires a higher update rate 54 compared to other regions around host vehicle 22. As an example, if the sensor is able to update the FOV at 10 scans/second, the region in ROI 24 will be scanned at 10 frames/second while the other portions of the FOV are scanned at 2 scans/second. This may greatly reduce the number of bits used for communication and processing.

Adjusting the update rate 54 of the sensors may be implemented in a variety of ways depending on the type of sensor. Two methods are described below. In a first approach, some sensors 36 will allow for the power to be turned off when certain portions of the FOV are scanned. For these sensors 36, controller 12 sends a signal to turn off sensors outside ROI 24 while keeping sensors inside ROI 24 on. For sensors 36 that cannot be turned off, a second method is required. In a second approach, the controller 12 selectively ignores the sensor detections in order to achieve the desired update rate 54. Controller 12 retains sensor information within ROI 24 and discards sensor information outside ROI 24. A combination of the first and second method is also possible, wherein the sensors 36 that can be controlled electrically are processed by the first method, while the other sensors 36 are processed by the second method. It should be noted that the first method is preferred because it saves power.

In addition to using map information to select ROI 24, road features 56 may be used to determine the type of sensor used within selected ROI 24. Typical examples of the object detector 20 may include a plurality of sensors 36, such as lidar, cameras, and radar. In one example, cameras have been used for lane marker detection. In some cases where lighting conditions are poor, such as in tunnels, lidar laser reflection may be used instead of a camera. In another example, radar is used for large-scale high-speed object 58 detection. In yet another example, lidar is used for pedestrian detection in urban areas because the lidar provides a large amount of detection compared to other sensors 36.

The pixel throughput or measurement that the lidar sensor in the object detector 20 can complete in a given time period is limited. By reducing update rate 54 outside ROI 24, more efficient measurements may be used at a given time or pixel throughput in ROI 24. One way to take advantage of the increased pixel throughput is to distribute it uniformly or non-uniformly in ROI 24 and produce a higher total pixel density, which means a higher angular resolution 48 in this region. For example, if ROI 24 is selected to be one-fourth the FOV of the sensor as shown in fig. 4A (1/4) and the update rate 54 of the sensor is reduced by one-third (1/3)42, then the pixel throughput in ROI 24 will be three times the raw throughput (3 x). The dot density in the scan lines can thus be increased to 3x if the line count in ROI 24 remains the same. In fig. 4B, the original pixel matrix is shown with solid dots 44 and the added pixels are shown with open dots 46.

Another way to take advantage of increased pixel throughput is to maintain the original scan image grid, but increase the signal-to-noise ratio (SNR) by averaging multiple measurements at the same point. This improvement in SNR is shown in fig. 5. The measurements before averaging are shown in FIG. 5A, butThe data after averaging is shown in fig. 5B, where the solid line 60 is the measured data and the dashed line 62 is the standard data. This increased SNR allows for the detection of weaker signals or returns from more distant objects while maintaining the original detection criteria. In the example shown in fig. 4A, a 3x pixel throughput would allow averaging of three measurements per image pixel while maintaining the raw image update rate 54 in ROI 24. In the case of averaging, the SNR will increase in magnitude by the square root of threeOr about 4.7dB, and the range will increase

For the same reason, by averaging multiple measurements at the same point, the SNR per pixel will improve. For targets at the same distance, a better SNR means a higher detection probability and a lower False Alarm Rate (FAR), or better image quality. In the example shown in FIG. 4A, the SNR will increase in magnitudeOr 4.7 dB. If the original SNR is 10dB, it is now 14.7dB in ROI 24.

The camera is an integral part of most object detectors 20. The camera is most widely used to cover the front side of the host vehicle 22, but it may also be used to cover the full 360 ° field of view 32 of the host vehicle 22. In one embodiment, the camera is reduced and enlarged to better match ROI 24. Adjusting the zoom information is a relatively simple operation and may be managed by the controller 12. The camera may also be rotated in the event that ROI 24 is located on the side of host vehicle 22 with insufficient camera coverage.

While the present invention has been described in terms of its preferred embodiments, it is not intended to be limited thereby, but rather only to the extent set forth in the following claims.