A kind of train obstacle detection method based on laser radar
阅读说明：本技术 一种基于激光雷达的列车障碍物探测方法 (A kind of train obstacle detection method based on laser radar ) 是由 周吉荣 韩毅 刘永涛 郭庆召 巨洪 于 2019-08-27 设计创作，主要内容包括：本发明公开了一种基于激光雷达的列车障碍探测方法。包含以下步骤：S01、获取障碍物激光雷达回波信号；S02、获取各个激光束对目标障碍物的投影点信息,得到点云数据；S03、对点云数据进行滤波处理；S04、列车轨道线识别；S05、点云数据聚类；S06、识别障碍物类型。本发明能够快速进行障碍物识别,采用建立激光雷达感兴趣区域与障碍物信息匹配的方式,大大避免了障碍物检测过程中繁琐的算法分析过程,有利于列车在高速工况下快速的检测列车前方障碍物。(The train obstacle detection method based on laser radar that the invention discloses a kind of.It comprises the steps of: S01, obtain barrier laser radar echo signal；S02, each laser beam is obtained to the subpoint information of target obstacle, obtain point cloud data；S03, point cloud data is filtered；S04, the identification of train rail line；S05, point cloud data cluster；S06, cognitive disorders species type.The present invention can quickly carry out obstacle recognition, using establishing laser radar area-of-interest and the matched mode of obstacle information, it largely avoided cumbersome algorithm analytic process during detection of obstacles, be conducive to train and quickly detect train front obstacle under high-speed working condition.)
1. a kind of train obstacle detection method based on laser radar, which is characterized in that comprise the steps of:
S01, barrier laser radar echo signal is obtained
Laser beam is issued by laser radar sensor, target area is detected, is come from by the reception of laser radar receiving unit The laser radar echo signal that target obstacle generates；
S02, point cloud data is obtained
Each laser beam is obtained to the subpoint information of target obstacle, obtains point cloud data；
S03, point cloud data filtering processing
Using radius filter method, point cloud data is filtered；
S04, the identification of train rail line
It establishes dynamic area-of-interest (ROI), analysis identification train rail line, dynamic realtime monitors train rail line；
S05, point cloud data cluster
Point cloud data cluster segmentation is carried out, obtains cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, and calculate relative position between target obstacle and train；
S06, cognitive disorders species type
By the feature extraction to cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, barrier profile information, the barrier profile information and obstacle that will acquire are obtained Object property data base is compared, can cognitive disorders species type.
2. a kind of train obstacle detection method based on laser radar according to claim 1, which is characterized in that In In S03, noise reduction process is carried out to point cloud data by radius filter method.
3. a kind of train obstacle detection method based on laser radar according to claim 1, which is characterized in that In In S04, according to the Z axis information of point cloud data, dynamic area-of-interest (ROI) is established, the laser reflection based on every a kind of medium Characteristic is different, and point cloud data obtained to laser radar carries out the classification of echo strength value filtering, then carries out train driving railway line Coarse extraction clusters railway line point cloud data using greatest hope clustering algorithm afterwards, then recycle least square method into The fitting of ranks track diatom.
4. a kind of train obstacle detection method based on laser radar according to claim 3, which is characterized in that In In S04, the train rail real-time monitoring is realized using Kalman filtering algorithm.
5. a kind of train obstacle detection method based on laser radar according to claim 1, which is characterized in that In In S05, different cluster threshold values are divided from laser radar relative distance according to laser site, are referred to using Euclidean distance as cluster Mark, is iterated cluster, final to obtain cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, and the y-axis of target obstacle and laser radar is obtained according to cluster point a small bundle of straw, etc. for silkworms to spin cocoons on Distance and elevation.
6. a kind of train obstacle detection method based on laser radar according to claim 5, which is characterized in that obtain Cluster point a small bundle of straw, etc. for silkworms to spin cocoons on the following steps are included:
Point cloud data in area-of-interest after step 1, reading noise reduction, randomly selects rmUnclassified random n point in space, Using K-D tree neighborhood search, mutual distance will be met in n vertex neighborhood less than threshold value dmPoint be divided into a kind of Pi。
Step 2, according to Euclidean distance calculation formula, fall in rmP in regioniClass and PjEuclidean distance between class are as follows:
Judge dij<dm, two classes are classified as one kind if being less than, if dij>dmThen the two is inhomogeneity.If there are still not in data The point of classification, then return to step 1)；
Step 3, after all the points are all sorted out, judge all kinds of midpoint quantity NiWith given threshold Nmin、NminIt is compared, If Ni<Nmin, then such is rejected；If Ni>NmaxSuch is then dismissed, is returned in step 1, d is accordingly reducedm, to the point data in such Iteration again is carried out, until all clusters all meet Nmin<Ni<NmaxWhen, end loop；
Target obstacle and the y-axis distance of laser radar and grid DEM formula are as follows:
7. a kind of train obstacle detection method based on laser radar according to claim 1, which is characterized in that In In S06, firstly, optimization cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, finds growth initial point, defines marginal point growth conditions, it is point-by-point to grow, find marginal point Collect V；Secondly, extract target obstacle profile with connected component labeling method, with well-established barrier property data base into Row Characteristic Contrast, so that it is determined that obstacle identity.
8. a kind of train obstacle detection method based on laser radar according to claim 1, which is characterized in that its mesh Mark barrier contours extract the following steps are included:
Step 1, selected growth initial point, all barrier point cloud datas are projected to XOZ plane, are calculated in XOZ plane Every point-to-point transmission mutual distance records mutual spacing maximum two o'clock ps、pe, two o'clock can be speculated as and be subordinated to edge point set V, and chosen Wherein it is used as initial point ps。
Step 2, search growth stepping point, define marginal point growth conditions, with psCentered on, establish ps-peVectorFor growth Direction,It can be consideredDelimit initial point withFor normal vector, radius is the semi-circular region of search of R, It is four direction by the region division, and searches for interior all neighborhood points.M(ps, ps+1) it is with psWith ps+1Straight line is axis, half Diameter is contained number in the cylindrical region of r, expression are as follows:
Therefore, selected stepping point ps+1It should meet:
In formula, { p1…pnBe same direction on point, MoFor threshold value of counting, G is growth step-length limit value.
Step 3, the point that each party is met to above-mentioned condition upwards are included in edge point set, choose any point as newly initial Point, and repeat step 1, step 2, until can not stepping, then chosen steps point concentrate in addition point as new stepping point；
Step 4, through the above steps 3, marginal points information orderly stores, and with connected component labeling method, extracts target barrier Hinder object profile.
Technical field
The present invention relates to train safe driving fields, and in particular to a kind of train obstacle detection based on laser radar Method.
Background technique
During train driving, the effective train front barrier that may be present that detects is for the driving safety of train Play the role of vital.And due to the particularity of train driving, train needs to detect barrier in remote situation In the presence of.Interference due to train driving people limited view or the visual field by other things, will lead to train driving people can not have The presence of the detection barrier of effect, at this stage there is many due to train can not effectively detect the presence of barrier and lead Train accident is caused, for the safety for ensureing train driving, the method for train obstacle detection merits attention.
Maximum braking distance following vehicle speed variation of the train in brake is bigger simultaneously, and train detects barrier Requirement just could must successfully guarantee that train being capable of safe in this presence apart from interior effective cognitive disorders object. In addition, there is many forms (such as: pedestrian, building steel tube, falling rocks) for train barrier, it is understood that there may be not dangerous factor Artificial obstacle, therefore the necessary identification for carrying out barrier when detecting train barrier, in order to avoid go out during train driving The case where existing barrier erroneous judgement, influence the normally travel of train.
The high speed of train driving proposes the requirement of high-timeliness for detection of obstacles, and current train obstacle Object detecting method is difficult to solve this problem.
Summary of the invention
In view of the above-mentioned problems of the prior art, proposing a kind of train obstacle detection based on laser radar in the present invention Method, improves train obstacle recognition ability, further ensures train safety traffic.
The technical scheme adopted by the invention is that: a kind of train obstacle detection method based on laser radar, comprising with Lower step:
S01, barrier laser radar echo signal is obtained
Laser beam is issued by laser radar sensor, target area is detected, is received by laser radar receiving unit The laser radar echo signal generated from target obstacle；
S02, point cloud data is obtained
Each laser beam is obtained to the subpoint information of target obstacle, obtains point cloud data；
S03, point cloud data filtering processing
Using radius filter method, point cloud data is filtered；
S04, the identification of train rail line
It establishes dynamic area-of-interest (ROI), analysis identification train rail line, dynamic realtime monitors train rail line；
S05, point cloud data cluster
Point cloud data cluster segmentation is carried out, obtains cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, and calculate relative position between target obstacle and train；
S06, cognitive disorders species type
By to cluster point a small bundle of straw, etc. for silkworms to spin cocoons on feature extraction, obtain barrier profile information, the barrier profile information that will acquire with Barrier property data base is compared, can cognitive disorders species type.
Preferably, in S03, noise reduction process is carried out to point cloud data by radius filter method.
Preferably, in S04, according to the Z axis information of point cloud data, dynamic area-of-interest (ROI) is established, based on each The Laser reflective characteristics of class medium are different, and point cloud data obtained to laser radar carries out the classification of echo strength value filtering, then carries out The coarse extraction of train driving railway line afterwards clusters railway line point cloud data using greatest hope clustering algorithm, then again The fitting of train rail line is carried out using least square method.
Preferably, in S04, the train rail real-time monitoring is realized using Kalman filtering algorithm.
Preferably, in S05, different cluster threshold values are divided from laser radar relative distance according to laser site, are used Euclidean distance is iterated cluster as clustering target, final to obtain cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, and obtains target disorders according to cluster point a small bundle of straw, etc. for silkworms to spin cocoons on The y-axis of object and laser radar distance and elevation.
Preferably, obtain cluster point a small bundle of straw, etc. for silkworms to spin cocoons on the following steps are included:
Point cloud data in area-of-interest after step 1, reading noise reduction, randomly selects rmUnclassified random n in space It is a, using K-D tree neighborhood search, mutual distance will be met in n vertex neighborhood less than threshold value dmPoint be divided into a kind of Pi。
Step 2, according to Euclidean distance calculation formula, fall in rmP in regioniClass and PjEuclidean distance between class Are as follows:
Judge dij< dm, two classes are classified as one kind if being less than, if dij> dmThen the two is inhomogeneity.If in data still There are unclassified points, then return to step 1)；
Step 3, after all the points are all sorted out, judge all kinds of midpoint quantity NiWith given threshold Nmin、NminIt carries out Compare, if Ni< Nmin, then such is rejected；If Ni> NmaxSuch is then dismissed, is returned in step 1, d is accordingly reducedm, in such Point data carry out iteration again, until all clusters all meet Nmin< Ni< NmaxWhen, end loop；
Target obstacle and the y-axis distance of laser radar and grid DEM formula are as follows:
Preferably, in S06, firstly, optimization cluster point a small bundle of straw, etc. for silkworms to spin cocoons on, finds growth initial point, defines marginal point growth conditions, by Point growth, finds edge point set V；Secondly, target obstacle profile is extracted, with well-established with connected component labeling method Barrier property data base carries out Characteristic Contrast, so that it is determined that obstacle identity.
Preferably, target obstacle contours extract the following steps are included:
Step 1, selected growth initial point, all barrier point cloud datas are projected to XOZ plane, in XOZ plane Every point-to-point transmission mutual distance is calculated, mutual spacing maximum two o'clock p is recordeds、pe, two o'clock can be speculated as and be subordinated to edge point set V, and It chooses and is wherein used as initial point ps。
Step 2, search growth stepping point, define marginal point growth conditions, with psCentered on, establish ps-peVector For the direction of growth,It can be consideredDelimit initial point withFor normal vector, radius is the semi-circular search of R The region division is four direction by region, and searches for interior all neighborhood points.M(ps, ps+1) it is with psWith ps+1During straight line is Axis, radius are contained number in the cylindrical region of r, expression are as follows:
Therefore, selected stepping point ps+1It should meet:
In formula, { p1...pnBe same direction on point, MoFor threshold value of counting, G is growth step-length limit value.
Step 3, the point that each party is met to above-mentioned condition upwards are included in edge point set, choose any point as at the beginning of new Initial point, and repeat step 1, step 2, until can not stepping, then chosen steps point concentrate in addition point as new stepping point；
Step 4, through the above steps 3, marginal points information orderly stores, and with connected component labeling method, extracts mesh Mark barrier profile.
It compares and the prior art, the invention has the benefit that
1, by establishing effective area-of-interest, and the obstacle of train driving interorbital is carried out by the way of characteristic matching Identification is surveyed in physical prospecting, be largely avoided cumbersome algorithm analytic process during detection of obstacles, is conducive to train in high-speed working condition Lower quickly detection train front obstacle.It solves since every frame of laser radar contains a data points up to ten thousand, is carrying out During analytical calculation, calculation amount is huge, and the processing time is easy the problem of extending.
2, initial point cloud data is filtered noise reduction process, and segment processing is carried out with the distance of point cloud data, Obstacle recognition error rate remotely is reduced, recognition efficiency is improved, recognition accuracy greatly improves.
Detailed description of the invention
Fig. 1 is invention working principle flow chart；
Fig. 2 is laser radar schematic view of the mounting position；
Fig. 3 is laser radar area-of-interest schematic diagram；
Fig. 4 is laser radar coordinate system schematic diagram.
Specific embodiment
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
A kind of train obstacle detection method based on laser radar, referring to Fig. 1, specific comprising hereinafter, step:
S01, barrier laser radar echo signal is obtained
- 上一篇：一种医用注射器针头装配设备
- 下一篇：一种基于航空遥感的输电走廊树障监测系统