A small obstacle detection method and device based on a laser radar point cloud

Abstract

The application provides a small obstacle detection method based on a laser radar point cloud, comprising: extracting gradient features and height features of the point cloud, performing a first segmentation based on the gradient features and the height features to preliminarily distinguish ground point clouds and non-ground point clouds; performing a second segmentation on the result of the first segmentation based on at least one of subdivision regions, intensity and density of the point cloud; performing clustering processing on the point cloud after the second segmentation; and estimating the shape and orientation of the small obstacle based on the result of the clustering processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of laser radar, and in particular to a small obstacle detection method based on laser radar point cloud. BACKGROUND

[0002] In the automatic driving of vehicles such as mine cards, cars, locomotives, and urban rails, the perception module of the vehicle serves as the eyes of the entire automatic driving system, providing reliable environmental information for the automatic driving system. In order to ensure the safety of the vehicle during driving, the perception module needs to accurately identify small obstacles in front of the vehicle, especially obstacles below 30cm*30cm*30cm, which can easily cause damage to the vehicle's tires and even cause tire blowout accidents, seriously endangering the safety of the vehicle and the passengers.

[0003] Under sufficient light conditions, the camera can detect the presence of small obstacles, but the distance accuracy of detecting small obstacles through the camera is limited, and the accurate position of the small obstacle cannot be accurately obtained. And in the case of insufficient light, the camera cannot effectively play a role. Therefore, in the automatic driving system, the perception of small obstacles relies more on high-resolution laser radars.

[0004] However, due to the influence of complex road conditions such as road undulations and slopes, general radar point cloud segmentation methods cannot effectively separate small obstacles from ground point clouds, which can easily cause false positives and false negatives, making it difficult for laser radars to accurately detect the shape and position of small obstacles, and more difficult to provide effective and accurate environmental information for vehicle driving, threatening the driving safety of automatic driving.

[0005] In order to overcome the above-mentioned defects existing in the prior art, there is an urgent need in the art for a small obstacle detection method based on laser radar point cloud, which is used for long-distance stable detection of small obstacles in complex scenes such as road undulations or unpaved roads, such as mine areas and rail transit scenes, so as to provide effective and accurate environmental information for vehicle driving, and ensure the driving safety of automatic driving. SUMMARY

[0006] The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

[0007] To overcome the above-mentioned defects existing in the prior art, the present application provides a small obstacle detection method based on a laser radar point cloud, comprising: extracting gradient features and height features of the point cloud, performing a first segmentation based on the gradient features and the height features to preliminarily distinguish ground point clouds and non-ground point clouds; performing a second segmentation on the result of the first segmentation based on at least one of the following: a subdivision region, intensity and density of the point cloud; performing clustering processing on the point cloud after the second segmentation; and estimating the shape and orientation of the small obstacle based on the result of the clustering processing.

[0008] In an embodiment, preferably, the second segmentation is performed based on the subdivision region of the point cloud, comprising: dividing the point cloud after the first segmentation into a plurality of subdivision regions, the plurality of subdivision regions comprising a near-ground region, an above-ground region and a below-ground region; and setting different gradient threshold values for different subdivision regions to perform the second segmentation.

[0009] In an embodiment, preferably, the second segmentation is performed based on the intensity of the point cloud, comprising: finding a seed point in the point cloud after the first segmentation according to the intensity of a single point and the average intensity of the surrounding neighborhood of the single point; and judging whether the gradient feature of the seed point meets a preset threshold requirement to perform the second segmentation.

[0010] In an embodiment, preferably, the second segmentation is performed based on the intensity of the point cloud, comprising: finding a seed point in the point cloud after the first segmentation according to the intensity of a single point and the average intensity of the surrounding neighborhood of the single point; and judging whether the gradient feature of the seed point meets a preset threshold requirement to perform the second segmentation.

[0011] In an embodiment, preferably, the second segmentation is performed based on the density of the point cloud, comprising: searching for the surrounding neighborhood of the ground point cloud after the first segmentation; and calculating the distance between points in the surrounding neighborhood of each ground point cloud, and clustering the point clouds whose distance between points is less than a preset threshold value into non-ground point clouds.

[0012] In an embodiment, preferably, before extracting the gradient features and the height features of the point cloud, the method further comprises: preprocessing the point cloud data; and extracting the gradient and height features of the preprocessed point cloud, the preprocessing of the point cloud data comprising: matrixing the point cloud data, the number of rows and the number of columns of the matrixing being determined by the following formula:

[0013]

[0014]

[0015] Wherein, VerAngleRange is the range of the vertical angle of the laser radar, VerResolution is the resolution of the vertical angle, rows is the number of rows of the matrix, HorAngleRange is the range of the horizontal angle of the laser radar, HorResolution is the resolution of the horizontal angle, and cols is the number of columns of the matrix.

[0016] In an embodiment, preferably, the pre-processing of the point cloud data further comprises: simultaneously increasing VerResolution and HorResolution in equal proportions to down-sample to improve operation efficiency.

[0017] In an embodiment, preferably, the gradient feature of the point cloud is extracted by: calculating the gradient of the point cloud P according to the following formula:

[0018]

[0019] Wherein, a is the gradient of the point cloud, M point and P point are respectively points in the point cloud depth map that are adjacent rows of the point cloud belonging to the same column, z p and z m are respectively the coordinates of M point and P point in the Z-axis direction, l p and l m are respectively the distances of the projection points of M point and P point in the XOY plane from the coordinate origin O.

[0020] In an embodiment, preferably, the height feature of the point cloud is extracted by: selecting fitting points according to the values in the Z direction of the point cloud by using a random sampling fitting algorithm; judging the degree of fluctuation of the ground according to the fitting points, obtaining a ground equation and calculating a segmentation threshold; and obtaining the height feature of each point from the ground according to the ground equation and the segmentation threshold to separate the ground point cloud and the non-ground point cloud.

[0021] In an embodiment, preferably, the clustering processing is performed on the point cloud after the second segmentation, comprising: clustering and dividing the non-ground point cloud after the second segmentation by using a density clustering algorithm to obtain a plurality of clustering categories.

[0022] In an embodiment, preferably, the shape and orientation of the small obstacle are estimated based on the result of the clustering processing, comprising: finding a minimum bounding box for the point cloud cluster of each clustering category, the minimum bounding box wrapping all the point clouds of the point cloud cluster; and determining the shape and orientation of the small obstacle based on the minimum bounding box.

[0023] In an embodiment, preferably, the minimum bounding box is a rectangular bounding box; and determining the shape and orientation of the small obstacle based on the minimum bounding box, comprising: the length and width of the rectangular bounding box are the length and width of the small obstacle, and the orientation of the rectangular bounding box is the orientation of the small obstacle.

[0024] Another aspect of the present application provides a small obstacle detection device based on a laser radar point cloud, comprising: a memory; and a processor coupled with the memory, the processor being configured to perform the steps of any of the above small obstacle detection methods.

[0025] The present application also provides a computer readable medium having stored thereon a computer program, the computer program being executed by a processor to implement the steps of any of the above small obstacle detection methods. BRIEF DESCRIPTION OF DRAWINGS

[0026] The above features and advantages of the present application will be better understood by reading the following detailed description of the embodiments of the present application in conjunction with the drawings, in which:

[0027] Figure 1 is a method flow diagram of a small obstacle detection method based on a laser radar point cloud according to an embodiment of the present application;

[0028] Figure 2 is a calculation principle diagram of a point cloud gradient feature according to an embodiment of the present application;

[0029] Figure 3 is a principle diagram of a minimum bounding box estimation method in clustering processing according to an embodiment of the present application; and

[0030] Figure 4 is a device structure diagram of a small obstacle detection device based on a laser radar point cloud according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] The specific embodiments of the present application will be described below by referring to the drawings, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Although the description of the present application will be introduced in combination with the preferred embodiments, this does not mean that the features of the present application are limited to the embodiments. On the contrary, the purpose of introducing the present application in combination with the embodiments is to cover other options or modifications that can be extended based on the claims of the present application. In order to provide a deep understanding of the present application, many specific details will be included in the following description. The present application can also be implemented without using these details. In addition, in order to avoid confusion or obscuring the focus of the present application, some specific details will be omitted in the description.

[0032] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0033] In addition, "up", "down", "left", "right", "top", "bottom", "horizontal", "vertical" used in the following description should be understood as the orientation shown in the section and the related drawings. The relative terms are only for the convenience of description, and do not mean that the device described thereby needs to be manufactured or operated in a particular orientation, and therefore should not be understood as a limitation on the present application.

[0034] It can be understood that although the terms "first", "second", "third" and the like are used herein to describe various components, regions, layers and / or parts, these components, regions, layers and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers and / or parts. Therefore, the first component, region, layer and / or part discussed below can be referred to as the second component, region, layer and / or part without departing from some embodiments of the present application.

[0035] In order to overcome the above-mentioned defects existing in the prior art, the present application provides a small obstacle detection method based on laser radar point cloud, which is used for stable detection of small obstacles in complex road conditions such as ground undulation and slope, so as to provide effective and accurate environmental information reference for vehicle driving, and ensure the driving safety of automatic driving.

[0036] Figure 1 The method flow diagram of the small obstacle detection method based on laser radar point cloud according to an embodiment of the present application is shown.

[0037] Please refer to Figure 1 The small obstacle detection method based on laser radar point cloud provided by the present application 100 comprises:

[0038] Step 101: Extract the gradient feature and height feature of the point cloud, and perform the first segmentation based on the gradient feature and height feature to preliminarily distinguish the ground point cloud and the non-ground point cloud.

[0039] Due to the unordered nature of point cloud data in space, in an embodiment, before extracting the gradient feature and height feature of the point cloud, the point cloud data is preprocessed, and the gradient and height feature of the preprocessed point cloud is extracted. The preprocessing of the point cloud data includes: matrixing the point cloud data, thereby facilitating subsequent segmentation and clustering operations on the point cloud.

[0040] The number of rows and columns of the matrix is determined by the following formula:

[0041]

[0042]

[0043] Wherein, VerAngleRange is the range of the vertical angle of the laser radar, VerResolution is the resolution of the vertical angle, rows is the number of rows of the matrix, HorAngleRange is the range of the horizontal angle of the laser radar, HorResolution is the resolution of the horizontal angle, and cols is the number of columns of the matrix.

[0044] In an embodiment, in order to improve the running efficiency of the algorithm and reduce the number of point searches, down-sampling is performed by increasing the values of VerResolution and HorResolution. When increasing the values of VerResolution and HorResolution, equal ratio increase is adopted to ensure that the shape of the obstacle profile is not deformed.

[0045] After the preprocessing of the point cloud data is completed, step 101 is performed: extracting the gradient feature and height feature of the point cloud, and performing the first segmentation based on the gradient feature and height feature to preliminarily distinguish the ground point cloud and the non-ground point cloud.

[0046] Figure 2 is a schematic diagram of the calculation principle of the gradient feature of the point cloud according to an embodiment of the present application.

[0047] After the point cloud is matrixed, the features between the beams are the features between the rows of the matrix, and the features within the beams are the features between the columns. The key to the segmentation of the ground point cloud and the non-ground point cloud is the change of the longitudinal gradient, as shown in Figure 2 The change of the gradient can be represented by the angle a.

[0048] Please refer to Figure 2 , M point and P point respectively correspond to two point clouds of adjacent rows in the same column in the point cloud depth map, z p and z m are the coordinates of the M point and the P point in the Z axis direction, l p and l m are the distances of the projection points of the M point and the P point in the XOY plane from the coordinate origin O.p With l m The formula is as follows:

[0049]

[0050]

[0051] Wherein, x p and y p are the X-axis coordinate and Y-axis coordinate of the point P respectively, x m and y m are the X-axis coordinate and Y-axis coordinate of the point M respectively.

[0052] The gradient of the point cloud is calculated by the following formula:

[0053]

[0054] Wherein, a is the gradient of the point cloud.

[0055] Because the angle a of the ground point cloud and the obstacle point cloud is quite different, the ground point cloud and the non-ground point cloud can be distinguished according to the gradient feature a of the point cloud. According to the experience value of actual debugging, the threshold a0 is taken, when a < a0 is satisfied, the point cloud in the current row and column in the depth map is judged as the ground point; otherwise, when a < a0 is not satisfied, it is judged as the non-ground point.

[0056] In addition to the gradient feature of the point cloud, the height feature of the point cloud also needs to be extracted. The height feature represents the height of each point from the ground, also known as the ground feature.

[0057] In an embodiment, the ground feature extraction process is realized by using the block plane fitting method. In the plane fitting process, the random sample consensus (RANSAC) algorithm is used.

[0058] The algorithm estimates the mathematical model from a set of observed data in an iterative manner. The RANSAC algorithm assumes that the data contains correct data and abnormal data. The correct data is called an inner point, and the abnormal data point is called an outer point. At the same time, the RANSAC algorithm also assumes that given a set of correct data, there is a method to calculate the model parameters that meet these data.

[0059] After the original point cloud is matrixed, the fitting points of RANSAC are selected according to the value of the point cloud in the z direction, the fluctuation degree of the ground is judged according to the total number of inner points and outer points, and the ground equation and the adaptive calculation of the ground segmentation threshold are obtained. According to the regional ground equation and the ground segmentation threshold, the non-ground points of each region can be separated, and the height feature of each point from the ground is obtained according to the point to plane equation, so that the ground point cloud and the non-ground point cloud are separated.

[0060] Please go back to Figure 1 The small obstacle detection method based on the laser radar point cloud 100 provided by the application further comprises:

[0061] Step 102: performing second segmentation on the result of the first segmentation based on at least one of the subdivided region, the intensity and the density of the point cloud.

[0062] The second segmentation is re-segmentation based on the first segmentation, and the point cloud of the small obstacle that is mis-segmented in the first segmentation is re-segmented, so that the number of point clouds retained by the small obstacle is more, and it is guaranteed that the small obstacle can be detected at a farther distance.

[0063] The twice segmentation of the ground point cloud and the non-ground point cloud in the small obstacle detection method based on the laser radar point cloud provided by the application comprises at least one of the following three: first, based on the features obtained by the first segmentation, the subdivided region of the point cloud is re-segmented by using different threshold values in the fixed region; second, the obstacle point cloud is extracted based on the single-point point cloud intensity and the gradient of the surrounding neighborhood; and third, the small obstacle point cloud is extracted based on the density features of the point cloud.

[0064] The second segmentation based on the subdivided region of the point cloud is that the point cloud after the first segmentation is divided into multiple subdivided regions, and the multiple subdivided regions include a near-ground region, a region above the ground and a region below the ground. By setting different gradient threshold values for point clouds in different regions, near-ground target point clouds are extracted more finely, and it is guaranteed that more point clouds of small targets can be segmented. The point clouds above or below the ground generally belong to larger obstacles, and a more relaxed threshold value can be used for segmentation, while a fine gradient threshold value is set in the near-ground region to perform the second segmentation.

[0065] In an embodiment, the gradient threshold value of the point cloud above or below the ground is selected as 0.8, and the gradient threshold value of the near-ground target point cloud is selected as 0.12. It can be understood that the gradient threshold value of the point cloud above or below the ground is selected to be more relaxed, and a larger value can be set, for example, in the range of 0.5 to 1. The gradient of the near-ground point cloud requires more fineness, and a smaller threshold value needs to be selected, for example, in the range of 0.1 to 0.5, so as to more finely and specifically divide the obstacles in different subdivided regions.

[0066] The point cloud intensity refers to the light intensity of a single point reflected by the laser radar. Generally, the point cloud intensity of the obstacle is significantly higher than that of the ground point cloud. Considering that the intensity information of the laser radar is not very accurate, the point cloud with prominent single-point intensity is considered, and the surrounding neighborhood is searched.

[0067] In an embodiment, the surrounding neighborhood of the point cloud is selected as a 5*5 25-neighborhood, and in actual working conditions, the neighborhood range can be flexibly selected according to different resolutions of the laser radar.

[0068] In an embodiment, the second segmentation is performed based on the intensity of the point cloud, including: in the ground point cloud after the first segmentation and the region segmentation, a seed point is found according to the average intensity of the surrounding neighborhood of the point cloud and the intensity of the single point. After the seed point is found, the gradient features of the points in the surrounding neighborhood thereof are searched, and it is judged whether the gradient features meet the threshold requirement, if the requirement is met, the region point cloud is segmented into non-ground point cloud, so that a more accurate segmentation result is obtained.

[0069] According to the scanning mode of the laser radar, the neighborhood point cloud of the non-ground point cloud is more dense. In an embodiment, the second segmentation is performed based on the density of the point cloud, including: the surrounding neighborhood of the ground point cloud after the first segmentation is searched, and the distance between the points in the surrounding neighborhood of each ground point cloud is calculated, if a point neighborhood contains multiple point clouds with close distances, the point cloud cluster is segmented into non-ground point cloud.

[0070] On the basis of the first point cloud segmentation, the second segmentation based on at least one of the subdivided regions, the intensity and the density of the point cloud is proposed, small obstacles can be more accurately and effectively identified, so as to adapt to the environmental requirements in complex road conditions such as ground undulation and slope.

[0071] Please go back to Figure 1 The small obstacle detection method based on the laser radar point cloud 100 provided by the application further includes:

[0072] Step 103: performing clustering processing on the point cloud after the second segmentation.

[0073] The clustering processing on the point cloud after the second segmentation is to divide the non-ground point cloud after the second segmentation into multiple clustering categories by using a density clustering algorithm.

[0074] After the ground segmentation and the interested region extraction, the non-ground point cloud in the interested region is obtained. In order to obtain obstacle information such as vehicles from the point cloud, the point cloud in the interested region needs to be further clustered. The fundamental purpose of the clustering algorithm is to divide the scattered laser radar point cloud into several independent point cloud sets. The obtained point cloud set is regarded as an obstacle.

[0075] In an embodiment, the clustering processing employs a DBSCAN algorithm. DBSCAN is a density-based clustering algorithm, which generally assumes that the density of the sample distribution can be determined. Samples of the same class are closely connected, so there must be samples of the same class around any sample of the class. By classifying closely connected samples as a class, a clustering class is obtained, and the number of obstacles and the corresponding point cloud cluster of the non-ground points are obtained.

[0076] Please go back to Figure 1 The laser radar point cloud-based small obstacle detection method 100 provided by the application further comprises:

[0077] Step 104: estimating the shape and orientation of the small obstacle based on the result of the clustering processing.

[0078] To obtain the state of the obstacle such as size and shape, shape estimation needs to be performed on the clustered point cloud.

[0079] In an embodiment, the shape and orientation of the small obstacle are estimated based on the result of the clustering processing, that is, a minimum bounding box is found for each point cloud cluster of the clustering class, and the minimum bounding box wraps all point clouds of the point cloud cluster; and the shape and orientation of the small obstacle are determined based on the minimum bounding box.

[0080] Figure 3 FIG. 1 is a schematic diagram of a principle of a minimum bounding box estimation method in clustering processing according to an embodiment of the application.

[0081] Please refer to Figure 3 The minimum bounding box estimation method finds, for each set of clustered laser point clouds, a minimum rectangular bounding box that can wrap all point clouds at a certain angle, and the possible direction θ of the rectangle ranges from 0° to 90°. Since the two adjacent sides of the rectangle are orthogonal, only a single side between 0° and 90° is considered, and the direction of the other side is θ+π / 2. The degree of fitting of the rectangular box is defined by the proximity of the clustered points to the two sides of the right angle:

[0082]

[0083] where i is defined as the i-th point in the m-point cluster, d i is the distance from the i-th point to the nearest side of the rectangular fitting box.

[0084] The sum of the reciprocals of the distances of all point clouds in the bounding box to the boundary is calculated, the angle is changed, and the above steps are repeated until all angles from 0° to 90° are traversed. The bounding box with the maximum reciprocal sum is the required minimum bounding box, and the length and width of the rectangular bounding box are the length and width of the small obstacle, and the orientation of the rectangular bounding box is the orientation of the small obstacle.

[0085] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0086] Figure 4 This is a schematic diagram of the device structure of a small obstacle detection device based on lidar point clouds according to an embodiment of the present invention.

[0087] like Figure 4 As shown, the computer system / server 400 of the small obstacle detection device is presented in the form of a general-purpose computer device. The components of the computer system / server 400 may include one or more processors 402, a memory 401, and a bus 403 connecting different system components (including the memory 401 and the processor 402).

[0088] Bus 403 includes a data bus, an address bus, and a control bus. The number of bits in the data bus is proportional to the data transfer rate (product of the operating frequency). The number of bits in the address bus determines the maximum addressable memory space. The control bus (read / write) indicates the type of bus cycle and the time when the current input / output operation is completed. Processor 402 is connected to memory 401 via bus 403 and configured to implement the vehicle control method provided in any of the above embodiments.

[0089] Processor 402, as the computing and control core of the computer system / server 400 of the small obstacle detection device, is the final execution unit for information processing and program execution. All software layer operations in the computer system are ultimately mapped to operations of processor 402 through the instruction set. The main functions of processor 402 are processing instructions, executing operations, controlling time, and processing data.

[0090] Memory 401 refers to various storage devices in a computer that store programs and data. Memory 401 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 404 and / or cache memory 405.

[0091] Random Access Memory (RAM) 404 is an internal memory that directly exchanges data with the processor 402. It can be read and written at any time (except during refresh) and is very fast. It is typically used as a temporary data storage medium for the operating system or other running programs. Data stored in it will be lost if power is lost. Cache Memory 405 is a level-one memory located between main memory and the processor 402. It has a smaller capacity but a much higher speed than main memory, approaching the speed of the processor 402.

[0092] The computer system / server 400 can also include other removable / non-removable, volatile / non-volatile computer system storage media. For example, storage system 406 can be used for storing data, programs and / or instructions. In one embodiment, storage system 406 can be a magnetic hard disk drive.

[0093] The storage system 406 can also include a removable / non-removable, volatile / non-volatile computer system storage medium. For example, storage system 406 can be implemented using a magnetic disk drive, a solid state drive, a floppy disk drive with suitable media, a hard disk drive with suitable media, or the like.

[0094] The computer system / server 400 can also communicate with one or more external devices 408 such as a keyboard, a pointing device, a display 409, etc.; one or more devices that enable a user to interact with computer system / server 400; and / or one or more devices that enable computer system / server 400 to communicate with one or more other computer systems or devices. Such communication can occur via Input / Output (I / O) interfaces 410.

[0095] The computer system / server 400 can communicate with one or more networks, such as a local area network (LAN), a general wide area network (WAN), and / or a public network such as the Internet, via network adapter 411. As depicted, network adapter 411 communicates with the other components of computer system / server 400 via bus 403. It should be understood that although not shown, other hardware and / or software components could be used in conjunction with computer system / server 400. Examples, include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc. Figure 4

[0096] The present application also provides an embodiment of a computer readable medium, and the computer storage medium stores a computer program. The computer program is executed by a processor, and can implement the steps of any one of the above-mentioned methods for detecting small obstacles based on a laser radar point cloud.

[0097] Those skilled in the art will understand that information, signals, and data can be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that can be referenced throughout the above description can be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof. ​

[0098] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans can implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

[0099] The processors described herein can be implemented using electronic hardware, computer software, or any combination thereof. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. By way of example, the processors presented in this disclosure, any portion of a processor, or any combination of processors presented in this disclosure can be implemented with a microprocessor, microcontroller, digital signal processor (DSP), field-programmable gate array (FPGA), programmable logic device (PLD), state machine, gated logic, discrete hardware circuits, and other suitable processing components configured to perform the various functions described throughout this disclosure. The functionality of the processors presented in this disclosure, any portion of a processor, or any combination of processors presented in this disclosure can be implemented with software executed by a microprocessor, microcontroller, DSP, or other suitable platform.

[0100] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as discrete components in a user terminal.

[0101] In one or more exemplary embodiments, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0102] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for small obstacle detection based on lidar point cloud, comprising: extracting gradient feature and height feature of the point cloud, performing first segmentation based on the gradient feature and the height feature to preliminarily distinguish ground point cloud and non-ground point cloud; performing second segmentation on the result of the first segmentation based on at least one of the following: subarea, intensity and density of the point cloud; performing clustering processing on the point cloud after the second segmentation; and estimating shape and position of the small obstacle based on the result of the clustering processing. performing second segmentation based on subarea of the point cloud, comprising: dividing the point cloud after the first segmentation into multiple subareas, the multiple subareas including near-ground area, above-ground area and below-ground area; and setting different gradient threshold values for different subareas to perform the second segmentation; performing second segmentation based on intensity of the point cloud, comprising: finding seed points in the point cloud after the first segmentation according to intensity of single point and average intensity of surrounding neighborhood of the single point; and judging whether gradient feature of the seed points meets preset threshold requirement to perform the second segmentation. the setting of different gradient threshold values for different subareas to perform the second segmentation, comprising:

2. The small obstacle detection method of claim 1, wherein, setting loose gradient threshold value in the above-ground area and the below-ground area to perform the second segmentation; and setting fine gradient threshold value in the near-ground area to perform the second segmentation. performing second segmentation based on density of the point cloud, comprising:

3. The small obstacle detection method of claim 1, wherein, searching surrounding neighborhood of ground point cloud after the first segmentation; and calculating distance between points in surrounding neighborhood of each ground point cloud, and clustering point cloud with distance between points less than preset threshold value as non-ground point cloud. before the extracting of gradient feature and height feature of the point cloud, further comprising:

4. The small obstacle detection method of claim 1, wherein, preprocessing point cloud data; and extracting gradient and height feature of the preprocessed point cloud, the preprocessing of point cloud data comprising: matrixing the point cloud data, number of rows and number of columns of the matrixing being determined by the following formula: the preprocessing of point cloud data, further comprising: wherein, is a range of the laser radar vertical angle, is a resolution of the vertical angle, is a number of rows of the matrix, is a range of the laser radar horizontal angle, is a resolution of the horizontal angle, is a number of columns of the matrix.

5. The small obstacle detection method of claim 4, wherein, the extracting of gradient feature of the point cloud, comprising: simultaneous uniform scaling and Downsampling is performed to improve computational efficiency.

6. The small obstacle detection method of claim 1, wherein, calculating gradient of the point cloud by the following formula: the extracting of the height feature of the point cloud, comprising: Wherein, a is the gradient of the point cloud, and M point and P point are respectively the point cloud of the adjacent rows in the point cloud depth map which belong to the same column with the point cloud, and are respectively the coordinates of M point and P point in the Z-axis direction, and are respectively the distance of the projection point of M point and P point in XOY plane from the coordinate origin O.

7. The small obstacle detection method of claim 1, wherein, selecting fitting points according to values in Z direction of the point cloud by random sampling fitting algorithm; judging fluctuation degree of ground according to the fitting points, obtaining ground equation and calculating segmentation threshold value; and obtaining height feature of each point from ground according to the ground equation and the segmentation threshold value to separate the ground point cloud and the non-ground point cloud. the performing of clustering processing on the point cloud after the second segmentation, comprising:

8. The small obstacle detection method of claim 1, wherein, clustering and dividing the non-ground point cloud after the second segmentation by density clustering algorithm to obtain multiple clustering categories. the estimating of shape and position of the small obstacle based on the result of the clustering processing, comprising:

9. The small obstacle detection method of claim 8, wherein, finding minimum bounding box for point cloud cluster of each clustering category, the minimum bounding box wrapping all point clouds of the point cloud cluster; and determining shape and position of the small obstacle based on the minimum bounding box. the minimum bounding box is rectangular bounding box.

10. The small obstacle detection method of claim 9, wherein, ​ The determining the shape and orientation of the small obstacle based on the minimum bounding box comprises: The length and width of the rectangular bounding box are the length and width of the small obstacle, and the orientation of the rectangular bounding box is the orientation of the small obstacle. 11.A small obstacle detection device based on laser radar point cloud, comprising: a memory; and a processor coupled to the memory, the processor being configured to perform the steps of the small obstacle detection method according to any one of claims 1-10.

12. A computer readable medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the small obstacle detection method according to any one of claims 1-10.

Images