Target detection method, device and equipment based on point cloud, medium and product

Abstract

The invention discloses a target detection method and device based on point cloud, equipment, a medium and a product, and relates to the field of target detection, and the method comprises the steps: obtaining point cloud data detected by a sensor; carrying out ROI region extraction and point cloud segmentation on the point cloud data to obtain a non-ground point cloud; performing point cloud detection by using an unsupervised clustering algorithm according to the non-ground point cloud to obtain a point cloud cluster; and according to the point cloud cluster, utilizing a deep network model to carry out prediction to obtain a category and a three-dimensional bounding box of the obstacle target. According to the invention, the precision of point cloud data target detection can be improved.

Description

Technical Field

[0001] The present application relates to the field of object detection, and in particular to an object detection method, device, equipment, medium and product based on point cloud. Background Art

[0002] With the rapid development of autonomous driving technology and advanced driver assistance technology, the accuracy of vehicle surrounding environment perception has become the key to ensuring driving safety. Cameras and lidar are common sensors in intelligent vehicle perception systems. Among them, cameras can provide rich color and texture information, but it is difficult to accurately obtain distance information and is easily affected by environmental conditions such as light changes. In contrast, lidar uses active emission and reception of laser beams to detect the surrounding environment, and has the advantages of accurate ranging and strong anti-interference ability. Therefore, obstacle detection based on lidar point cloud data has become a research hotspot.

[0003] Object detection algorithms based on point cloud can be divided into deep learning-based methods and clustering-based methods. Deep learning-based methods can achieve accurate detection of obstacles. However, deep learning-based detection methods have high requirements for the quality of training data annotation, can only recognize the target categories included in the training set, and there are missed detections for obstacles of untrained types, which easily lead to recognition errors and even traffic accidents. Traditional clustering-based obstacle recognition methods do not require training and can achieve obstacle detection to a certain extent. However, due to the sparsity of lidar point cloud data and the influence of obstacle angles and positions, it is difficult to completely perceive obstacles, so their sizes cannot be accurately described, increasing the accident risk. Therefore, how to improve the accuracy and reliability of obstacle detection in point cloud data in complex driving scenarios has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of the present application is to provide an object detection method, device, equipment, medium and product based on point cloud, which can improve the accuracy of object detection in point cloud data.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In the first aspect, the present application provides an object detection method based on point cloud, including:

[0007] Obtain point cloud data detected by a sensor;

[0008] Extract the ROI region and segment the point cloud from the point cloud data to obtain non-ground point cloud;

[0009] Perform point cloud detection on the non-ground point cloud using an unsupervised clustering algorithm to obtain point cloud clusters;

[0010] Predict using the depth network model based on the point cloud cluster to obtain the category and three-dimensional bounding box of the obstacle target.

[0011] Optionally, before performing ROI region extraction and point cloud segmentation on the point cloud data to obtain non-ground point clouds, it further includes:

[0012] Perform coordinate transformation on the point cloud data detected by the sensor using a coordinate transformation matrix, where the coordinate transformation matrix is the coordinate transformation matrix between the vehicle body coordinate system and the sensor coordinate system.

[0013] Optionally, performing ROI region extraction and point cloud segmentation on the point cloud data to obtain non-ground point clouds specifically includes:

[0014] Set the ROI region range of the point cloud according to a preset aiming position;

[0015] Extract the point cloud in the ROI region from the point cloud data according to the ROI range region to obtain the ROI region point cloud;

[0016] Use a point cloud segmentation algorithm to segment the ROI region point cloud to obtain non-ground point clouds.

[0017] Optionally, using a point cloud segmentation algorithm to segment the ROI region point cloud to obtain non-ground point clouds specifically includes:

[0018] Partition the ROI region point cloud to obtain the partitioned point cloud;

[0019] Allocate the partitioned point cloud to grids;

[0020] Use a plane fitting method to segment the point cloud in each grid to obtain non-ground point clouds.

[0021] Optionally, the plane fitting method includes the random sample consensus method and the principal component analysis method.

[0022] Optionally, the unsupervised clustering algorithm includes the clustering algorithm based on Euclidean distance and the density-based clustering algorithm.

[0023] In a second aspect, the present application provides a point cloud-based target detection device, including:

[0024] An acquisition module for acquiring the point cloud data detected by the sensor;

[0025] An extraction and segmentation module for performing ROI region extraction and point cloud segmentation on the point cloud data to obtain non-ground point clouds;

[0026] A point cloud detection module for performing point cloud detection on the non-ground point cloud using an unsupervised clustering algorithm to obtain a point cloud cluster;

[0027] A prediction module, configured to perform prediction using a deep network model based on the point cloud clusters to obtain the category and three-dimensional bounding box of the obstacle target.

[0028] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned point cloud-based target detection method.

[0029] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned point cloud-based target detection method.

[0030] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above-mentioned point cloud-based target detection method.

[0031] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0032] The present application provides a point cloud-based target detection method, device, equipment, medium, and product. First, the ground point cloud is filtered out by using ROI region extraction and point cloud segmentation to improve data accuracy, and an unsupervised clustering algorithm is used to perform point cloud detection on the non-ground point cloud to remove point cloud noise, further improving the accuracy of target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the following described drawings are only some embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0034] Figure 1 It is an application environment diagram of a point cloud-based target detection method in an embodiment of the present application;

[0035] Figure 2 It is a flowchart of a point cloud-based target detection method provided in an embodiment of the present application;

[0036] Figure 3 It is a schematic diagram of a point cloud-based target detection method provided in an embodiment of the present application;

[0037] Figure 4 It is a block diagram of a point cloud-based target detection method provided in an embodiment of the present application;

[0038] Figure 5 Schematic diagram of a computer device provided by an embodiment of the present application. Specific implementation manners

[0039] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0040] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0041] Body coordinate system B: The body coordinate system is fixedly connected to the vehicle platform, and the origin O vehicle is the center of mass of the vehicle platform. The x-axis direction is defined as the forward direction of the vehicle platform, the y-axis direction is defined as the left side direction of the vehicle platform, and the z-axis direction is defined as vertically upward.

[0042] Sensor coordinate system S: The sensor coordinate system is fixedly connected to the sensor, and the origin O sensor is the geometric center of the sensor, and the coordinate axis directions are the same as those of the body coordinate system.

[0043] Typical category of obstacles: Obstacles of the types included in the training dataset. (Including but not limited to common vehicles, pedestrians, etc. in the driving scenario).

[0044] General obstacles: Non-typical obstacles in the driving scenario, obstacles of types not included in the training set (including but not limited to obstacles of irregular sizes, roadblocks, trees, etc.).

[0045] Three-dimensional bounding box: Used to describe the position and size information of an object in three-dimensional space, represented by the center point coordinates (Center_x, Center_y, Center_z) of the bounding box, the three-dimensional size (l, w, h) of the bounding box, and the yaw angle.

[0046] Traditional clustering-based point cloud object detection methods belong to unsupervised learning methods and can detect obstacles in the scene without training. However, due to the sparsity of point cloud data, some areas of the obstacles may not be detected by the lidar, resulting in incomplete perception of the obstacles. This makes it difficult for clustering algorithms to accurately and comprehensively describe the position and size information of the obstacles, thereby affecting the effectiveness of the subsequent path planning module and increasing the risk of accidents.

[0047] Deep learning-based point cloud detection methods require a large amount of high-quality labeled data to train deep network models. However, the transferability of these methods is poor, and when adapting to different types of lidar, performance degradation may occur. In addition, due to the limited generalization ability of deep learning models, it is difficult to detect target categories not included in the training set, which easily leads to missed detections or false detections, thus increasing the safety risk of vehicle driving.

[0048] The method based on the fusion of deep learning models and unsupervised clustering combines the high-precision characteristics of deep learning methods and the flexibility of unsupervised clustering methods, making up for the deficiencies of single methods. However, the fusion method needs to effectively integrate the outputs of the two methods, and complex fusion strategies need to be designed, increasing the complexity of algorithm design and the difficulty of deployment and development.

[0049] In view of the deficiencies of the above methods, in order to accurately identify and describe the obstacles in the vehicle's surrounding environment, this application proposes a point cloud-based target detection algorithm. This method first preprocesses the environmental point cloud data captured by sensors (including but not limited to lidar, binocular cameras, etc.), and extracts the point cloud in the RoI (Region of Interest) area; then, a point cloud segmentation model is constructed to remove the ground point cloud data in the RoI area, and the non-ground point cloud data is clustered (including but not limited to: density-based clustering (DBSCAN), density-based clustering and other methods) to obtain the point cloud clusters of obstacles in the environment, and the three-dimensional bounding box of the point cloud clusters is initially calculated; a target feature recognition network based on the point cloud clusters is constructed to detect the clustered point cloud clusters, and realize the prediction of the types of typical category point cloud clusters (including but not limited to people, vehicles, etc.) and the update of features such as three-dimensional bounding boxes.

[0050] In summary, the target detection algorithm proposed in this application can accurately identify the typical category obstacles included in the dataset through ground point cloud removal, clustering segmentation, and deep learning-based target feature recognition, and can also effectively detect general obstacles in the driving scenario. This algorithm realizes the accurate identification and comprehensive description of obstacles in the vehicle's surrounding environment, provides a more reliable environment perception ability for the autonomous driving system, and thus significantly improves driving safety.

[0051] The point cloud-based target detection method provided by the embodiments of this application can be applied to, for example Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set separately, integrated on the server 104, placed on the cloud or other servers. The terminal 102 can send the point cloud data to the server 104. After receiving the point cloud data, for the point cloud data, the server 104 extracts the ROI region and performs point cloud segmentation on the point cloud data to obtain non-ground point clouds; performs point cloud detection on the non-ground point clouds using an unsupervised clustering algorithm to obtain point cloud clusters; and uses a deep network model to predict based on the point cloud clusters to obtain the category and 3D bounding box of the obstacle target. The server 104 can feedback the obtained category and 3D bounding box of the obstacle target to the terminal 102. In addition, in some embodiments, the object detection method based on point clouds can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly perform object detection on the point cloud data, or the server 104 can obtain the point cloud data from the data storage system and perform object detection on the point cloud data.

[0052] Among them, the terminal 102 can be, but is not limited to, various intelligent vehicle-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0053] In an exemplary embodiment, such as Figure 2 、 Figure 3 and Figure 4 shown, a method for object detection based on point clouds is provided. This method is executed by a computer device, and can be specifically executed alone by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, taking this method applied to Figure 1 the server 104 in

[0054] as an example, it includes the following steps 201 to step 204.

[0055] Among them:

[0056] Step 201: Obtain the point cloud data detected by the sensor.

[0057] Step 202: Extract the ROI region and perform point cloud segmentation on the point cloud data to obtain non-ground point clouds.

[0058] Step 203: Perform point cloud detection on the non-ground point clouds using an unsupervised clustering algorithm to obtain point cloud clusters.

[0059] Performing the above-mentioned steps 201 to 204 can improve the accuracy of object detection.

[0060] In an exemplary embodiment of the present application, before performing ROI region extraction and point cloud segmentation on the point cloud data to obtain non-ground point cloud, it further includes:

[0061] Performing coordinate transformation on the point cloud data detected by the sensor using a coordinate transformation matrix, where the coordinate transformation matrix is the coordinate transformation matrix between the vehicle body coordinate system and the sensor coordinate system.

[0062] In practical applications, performing coordinate transformation is for sensor joint calibration. Among them, the sensor joint calibration part includes sensor-vehicle platform joint calibration to determine the transformation relationship between the sensor coordinate system S and the vehicle body coordinate system B.

[0063] Through pre-offline calibration, the coordinate transformation matrix between the vehicle body coordinate system B and the sensor coordinate system S is known:

[0064]

[0065] Among them, R BS represents the rotation matrix from the vehicle body coordinate system B to the sensor coordinate system S, with a matrix size of 3×3, and t BS represents the translation matrix from the vehicle body coordinate system B to the sensor coordinate system S, with a matrix size of 3×1, and T BS is the coordinate transformation matrix between the vehicle body coordinate system B and the sensor coordinate system S, and 0 1×3 is the extension of the augmented matrix.

[0066] In an exemplary embodiment of the present application, performing ROI region extraction and point cloud segmentation on the point cloud data to obtain non-ground point cloud specifically includes: setting the ROI region range of the point cloud according to a preset aiming position; performing point cloud extraction on the point cloud data according to the ROI range region to obtain ROI region point cloud; using a point cloud segmentation algorithm to segment the ROI region point cloud to obtain non-ground point cloud.

[0067] In an exemplary embodiment of the present application, using a point cloud segmentation algorithm to segment the ROI region point cloud to obtain non-ground point cloud specifically includes: partitioning the ROI region point cloud to obtain partitioned point cloud; allocating the partitioned point cloud to grids; using a plane fitting method to segment the point cloud in each grid to obtain non-ground point cloud.

[0068] The process of obtaining non-ground point clouds is to remove ground point clouds. Among them, the part of removing ground point clouds includes point cloud preprocessing and point cloud segmentation. According to the preset preview distance, set the ROI area range of the point cloud and extract the point cloud of the ROI. Use the point cloud segmentation algorithm to segment the point cloud in the ROI area into ground point clouds and non-ground point clouds.

[0069] In practical applications, before removing the ground point cloud, first transform the point cloud data obtained in the sensor coordinate system S to the vehicle body coordinate system B through the coordinate transformation matrix. According to the preset detection range, extract the point cloud within the region of interest (ROI) from the point cloud data, and further analyze and process the point cloud data within a specific region range to reduce the interference of the point cloud in the non-ROI region and improve the calculation speed.

[0070] In the process of obtaining environmental point clouds, a sensor is used to scan the vehicle's surrounding environment, and the obtained point cloud data covers ground area point clouds and non-ground area point clouds. In order to detect obstacle targets in the environment, it is necessary to remove the ground area point clouds. In actual road scenarios, situations such as road gradients, uneven road surfaces, and road discontinuities may lead to problems of over-segmentation or under-segmentation when directly segmenting the global environmental point cloud. To cope with complex road scenarios, based on the assumption that the local road area is flat, the environmental point cloud is partitioned and assigned to grids of equal size, and a plane fitting method (including but not limited to: the method based on random sample consensus, the method based on principal component analysis, etc.) is used to segment the point cloud within each sub-grid. This local segmentation method helps reduce the risk of mis-segmentation, improve the accuracy of the ground area point cloud, remove the ground point cloud within each sub-grid area, and retain the non-ground point cloud data. Specifically, first extract the point cloud data within the region of interest, then divide it into the grid model according to the coordinates of the point cloud, perform plane fitting on the point cloud within each grid area, first segment the point cloud within the grid, and fuse the point cloud data obtained by segmentation within each grid.

[0071] In an exemplary embodiment of the present application, the plane fitting method includes the random sample consensus method and the principal component analysis method.

[0072] In an exemplary embodiment of the present application, step 203 is to perform unsupervised clustering detection. Among them, the part of unsupervised clustering detection includes point cloud clustering and point cloud cluster fusion. Traverse the non-ground point cloud, and the unsupervised clustering algorithm realizes the clustering recognition of the point cloud according to the distance relationship between the point clouds. Calculate the distance information between the point cloud clusters, and merge the point cloud clusters that meet the distance requirements to avoid misdetecting a large-size obstacle as multiple obstacles.

[0073] In practical applications, based on the non-ground point cloud obtained after ground removal, an unsupervised clustering algorithm (including but not limited to clustering algorithms based on Euclidean distance, density-based clustering algorithms, etc.) is used to detect the point cloud. Taking the DBSCAN algorithm as an example below:

[0074] 1. Initialize clustering parameters: Define the search neighborhood ε and the minimum neighborhood points MinPts.

[0075] 2. Traverse all points and classify each point in the data into three main categories:

[0076] Core point: If the neighborhood within ε of a point contains at least MinPts points (including itself), then this point is marked as a core point.

[0077] Border point: If the neighborhood within ε of a point contains fewer than MinPts points, but this point is within the neighborhood of a core point, it is marked as a border point.

[0078] Noise point: If a point is neither a core point nor a border point, then this point is a noise point.

[0079] 3. For each core point, expand the neighborhood points and form clusters.

[0080] Process border points and noise points, add border points to the appropriate clusters, and noise points do not belong to any cluster.

[0081] 4. Output the clustering results, including each cluster and noise points.

[0082] According to the output of the DBSCAN algorithm, remove the point cloud belonging to noise points, remove the noise points after clustering, retain the point cloud clusters obtained by clustering, and calculate the center coordinates (C x , C y , C z ) of the point cloud clusters.

[0083]

[0084] where n is the number of points in the point cloud cluster, and (x i , y i , z i ) is the coordinate of the i-th point among them.

[0085] Set the distance threshold l, traverse the distances between the center coordinates of the point cloud clusters. If the distance between the center coordinates is less than the distance threshold l, then merge the corresponding point cloud clusters and update the center coordinates of the point cloud clusters.

[0086] Obtain the final point cloud clusters and calculate the three-dimensional bounding box of the point cloud based on the preliminary calculation of the point cloud clusters.

[0087] In an exemplary embodiment of the present application, the unsupervised clustering algorithm includes a clustering algorithm based on Euclidean distance and a density-based clustering algorithm.

[0088] In an exemplary embodiment of the present application, step 204 is the detection process of the deep network model, and the deep network model detection part includes an offline training stage and an online detection stage. In the offline training stage, a deep neural network structure is built, a loss function is designed, and the model is trained based on a public dataset. The input of the model is the point cloud clusters obtained by clustering, and the output is features such as the category of the target and the three-dimensional bounding box. In the online detection stage, the point cloud data obtained by real-time clustering is input into the model obtained by offline training, and information such as the category, position, and size of the obstacles around the vehicle is output.

[0089] In practical applications, in the offline stage, first build a deep network model (such as a convolutional neural network, a fully connected multi-layer perceptron, etc.), and train it using the labeled data in the public dataset. The input of the model is the point cloud clusters, and the output is the category of the target and the three-dimensional bounding box information. In the online detection stage, input the point cloud clusters obtained in the previous step into the network model trained offline to realize the category recognition of typical category obstacles and the prediction of their three-dimensional bounding boxes, and update the three-dimensional bounding boxes calculated before. For general obstacles of unknown types, no update processing is performed.

[0090] The object detection algorithm proposed in the present application combines the advantages of traditional unsupervised clustering methods and deep learning techniques. In the first stage, the preliminary detection of obstacles in the environment is realized through the unsupervised clustering method, which can efficiently extract potential targets from the point cloud data without relying on labeled data; in the second stage, the deep learning model is used to further accurately identify the point cloud clusters and predict the category, three-dimensional bounding box and other key features of the target. This fusion strategy ensures the robustness of the algorithm under different environmental conditions, can adapt to various complex scenarios, and reduces the occurrence of false detections and missed detections. At the same time, by combining the advantages of the two methods, the model can not only accurately identify the types of surrounding obstacles, but also comprehensively describe the spatial distribution and dynamic characteristics of the targets, providing more comprehensive and accurate environmental perception data support for the advanced driver assistance system (ADAS) of intelligent vehicles, thereby improving the safety, stability and adaptability of the autonomous driving system.

[0091] Based on the same inventive concept, the embodiment of the present application also provides a point cloud-based object detection device for implementing the above-mentioned point cloud-based object detection method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the point cloud-based object detection device provided below can refer to the limitations on the point cloud-based object detection method in the above text, and will not be repeated here.

[0092] In an exemplary embodiment, a point cloud-based target detection device is provided, including:

[0093] An acquisition module, configured to acquire point cloud data detected by a sensor;

[0094] An extraction and segmentation module, configured to perform ROI region extraction and point cloud segmentation on the point cloud data to obtain non-ground point clouds;

[0095] A point cloud detection module, configured to perform point cloud detection on the non-ground point clouds using an unsupervised clustering algorithm to obtain point cloud clusters;

[0096] A prediction module, configured to perform prediction on the point cloud clusters using a deep network model to obtain the category and 3D bounding box of an obstacle target.

[0097] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal, and its internal structure diagram may be as shown in Figure 5 The figure shows. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the category and 3D bounding box data of obstacle targets. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a point cloud-based target detection method is implemented.

[0098] Those skilled in the art can understand that Figure 5 The structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different component layout. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the above method embodiments are implemented.

[0099] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program which, when executed by a processor, implements the above-described method embodiments.

[0100] In an exemplary embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the above-described method embodiments.

[0101] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0102] In this application, all actions of obtaining signals, information, or data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where it is located and obtaining authorization from the owner of the corresponding device.

[0103] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above-described method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-described method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memories (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0104] In each of the embodiments provided in this application, the database involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., without limitation. In each of the embodiments provided in this application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without limitation.

[0105] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0106] Specific examples are used in this article to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A point cloud-based target detection method, characterized in that: The point cloud-based target detection method comprises: Obtain point cloud data detected by the sensor; Performing ROI region extraction and point cloud segmentation on the point cloud data to obtain a non-ground point cloud; Performing point cloud detection using an unsupervised clustering algorithm according to the non-ground point cloud to obtain point cloud clusters; The deep network model is used to perform prediction based on the point cloud cluster to obtain the category and three-dimensional bounding box of the obstacle target.

2. The point cloud-based target detection method according to claim 1, characterized in that: Before performing ROI region extraction and point cloud segmentation on the point cloud data to obtain a non-ground point cloud, the method further includes: The point cloud data detected by the sensor is subjected to coordinate transformation using a coordinate transformation matrix, wherein the coordinate transformation matrix is ​​a coordinate transformation matrix of a vehicle body coordinate system and a sensor coordinate system.

3. The point cloud-based target detection method according to claim 1, characterized in that: Performing ROI region extraction and point cloud segmentation on the point cloud data to obtain a non-ground point cloud specifically includes: Set the ROI area range of the point cloud according to the preset aiming position; Extracting point cloud from the point cloud data according to the ROI range area to obtain a point cloud in the ROI area; The point cloud in the ROI area is segmented using a point cloud segmentation algorithm to obtain a non-ground point cloud.

4. The point cloud-based target detection method according to claim 3, characterized in that: The point cloud in the ROI area is segmented using a point cloud segmentation algorithm to obtain a non-ground point cloud, specifically including: Partitioning the point cloud in the ROI area to obtain a partitioned point cloud; Allocating the partitioned point cloud to a grid; Based on the road surface setting condition, the point cloud in each grid is segmented by using a plane fitting method to obtain a non-ground point cloud; the road surface setting condition is that the road surface in the local area is flat.

5. The point cloud-based target detection method according to claim 4, characterized in that: The plane fitting method includes a random sampling consistency method and a principal component analysis method.

6. The point cloud-based target detection method according to claim 1, characterized in that: The unsupervised clustering algorithm includes a clustering algorithm based on Euclidean distance and a clustering algorithm based on density.

7. A point cloud-based target detection device, characterized in that: The point cloud-based target detection device comprises: An acquisition module is used to acquire point cloud data detected by the sensor; An extraction and segmentation module, used to perform ROI region extraction and point cloud segmentation on the point cloud data to obtain a non-ground point cloud; A point cloud detection module, used to perform point cloud detection based on the non-ground point cloud using an unsupervised clustering algorithm to obtain a point cloud cluster; The prediction module is used to predict the category and three-dimensional bounding box of the obstacle target by using a deep network model according to the point cloud cluster.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the point cloud-based target detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the point cloud-based target detection method described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the point cloud-based target detection method described in any one of claims 1 to 6 is implemented.

Images