Target detection method and device, computer equipment and storage medium

Abstract

The invention relates to a target detection method and device, computer equipment and a storage medium. The method belongs to the technical field of intelligent driving, and comprises the following steps: carrying out segmentation processing on point cloud data of a target object to obtain at least two original point cloud columns associated with the target object; and based on the central position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, determining the target feature of the point cloud contained in each original point cloud column. And detecting a target object based on the target features of the original point cloud columns and the point clouds contained in the original point cloud columns. Based on the target detection method provided by the invention, high-precision detection of the target object can be realized, and the problem of inaccurate detection result of the target object with a small volume in the prior art is effectively solved.

Description

Technical Field

[0001] The present application relates to the field of intelligent driving technology, and in particular to a target detection method, device, computer equipment and storage medium. Background Art

[0002] With the rapid development of artificial intelligence technology, many vehicles have been equipped with autonomous driving functions. Among them, the environmental perception module, as the first link of the intelligent driving system, needs to provide accurate information about the vehicle's surrounding environment for subsequent decision-making planning and motion control as a basis. It is an important foundation and core technology of the autonomous driving system.

[0003] The key to the environmental perception module is the 3D target detection algorithm, which uses sensors such as lidar to collect data information about obstacles such as vehicles and pedestrians in three-dimensional space. However, in high-speed scenes of autonomous driving, the detection algorithm for small target obstacles at long distances is still extremely challenging. How to improve the detection accuracy of small target obstacles is still an important problem that needs to be solved urgently. Summary of the invention

[0004] Based on this, it is necessary to provide a target detection method, device, computer equipment and storage medium that can effectively improve the engine braking effect in response to the above technical problems.

[0005] In a first aspect, the present application provides a target detection method. The method comprises:

[0006] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0007] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0008] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0009] In one embodiment, determining target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column includes:

[0010] For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column;

[0011] According to the center position, and the original position and relative position of the point cloud contained in the original point cloud column, the information features of the point cloud contained in the original point cloud column are determined;

[0012] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

[0013] In one embodiment, the attention model includes a point attention model and a point-column attention model;

[0014] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column, including:

[0015] Input the information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column;

[0016] The point column attention model is used to splice the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain the splicing features of the point cloud;

[0017] The attention weighted features and the concatenated features are multiplied to obtain the target features of the point cloud.

[0018] In one embodiment, the dot-column attention model includes a maximum pooling layer, an average pooling layer, and a fully connected layer;

[0019] The information features of the point cloud contained in the original point cloud column are input into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column, including:

[0020] Input the information features of the point cloud contained in the original point cloud column into the maximum pooling layer, and obtain the first descriptor output by the maximum pooling layer after pooling the information features of the point cloud;

[0021] Input the information features of the point cloud contained in the original point cloud column into the average pooling layer, and obtain the second descriptor output by the average pooling layer after performing pooling processing on the information features of the point cloud;

[0022] The first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature, and outputs the attention weighted feature of the point cloud.

[0023] In one embodiment, detecting the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column includes:

[0024] Determine a pseudo image of the target object based on each original point cloud column and target features of the point cloud contained in each original point cloud column;

[0025] The pseudo image is input into the target detection model, and the target detection model detects the target object and outputs the detection result.

[0026] In one embodiment, the method further comprises:

[0027] For each original point cloud column, determine the geometric center of all point clouds contained in the original point cloud column according to the original positions of the point clouds contained in the original point cloud column;

[0028] The center position corresponding to the geometric center is used as the center position of the original point cloud column.

[0029] In a second aspect, the present application also provides a target detection device. The device comprises:

[0030] A segmentation module, used to segment the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0031] A first determination module, for determining target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0032] The detection module is used to detect the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0033] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program.

[0034] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0035] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0036] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0038] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0039] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0040] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0041] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0042] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0043] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0044] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0045] The above-mentioned target detection method, device, computer equipment and storage medium segment and process the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object. Based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, the target features of the point cloud contained in each original point cloud column are determined. Based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected. Based on the target detection method of the present application, high-precision detection of the target object can be achieved, which effectively solves the problem of inaccurate detection results for smaller targets in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] FIG1 is a diagram showing an application environment of a target detection method provided by this embodiment;

[0047] FIG2 is a schematic diagram of a flow chart of a first target detection method provided by this embodiment;

[0048] Figure 3 A schematic diagram of a flow chart for determining a target valve clearance of an engine provided in this embodiment;

[0049] Figure 4 A schematic diagram of a flow chart for adjusting the engine valve clearance provided in this embodiment;

[0050] Figure 5 A schematic diagram of a flow chart of a second target detection method provided in this embodiment;

[0051] Figure 6 A structural block diagram of a target detection device provided in this embodiment;

[0052] FIG. 7 is a diagram showing the internal structure of the computer device provided in this embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] The vehicle state analysis method provided in the embodiment of the present application can be executed by a vehicle-mounted terminal, or can be executed by a server. That is, the method can be applied to the application environment shown in Figure 1. Take the application to the server as an example for illustration, wherein the server 104 obtains the point cloud data for the target object uploaded by the target vehicle 102. The server 104 performs segmentation processing on the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object. Based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, the target features of the point cloud contained in each original point cloud column are determined. The server 104 detects the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0055] The server 104 may be implemented as an independent server or a server cluster composed of multiple servers. The target vehicle may be a fuel vehicle or a new energy vehicle, and the target vehicle should have the ability to upload vehicle operation information to the Internet of Vehicles platform.

[0056] In one embodiment, as shown in FIG. 2 , a target detection method is provided, and the method is applied to the server in FIG. 1 as an example for explanation. As shown in FIG. 2 , the method includes the following steps:

[0057] S201, segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object.

[0058] Among them, the target object refers to the target object that needs to be detected, which can be a person or other obstacles. Point cloud refers to a data representation method that represents the surface of an object or scene through three-dimensional coordinate points. It is a visualization form of spatial data, consisting of a large number of three-dimensional coordinate points, each of which represents a position in space, and usually includes other additional information, such as color, intensity, texture, etc. Point cloud column refers to the point cloud data structure represented in a columnar form in three-dimensional point cloud data. The original point cloud column refers to the point cloud column obtained after the point cloud data of the target object is segmented.

[0059] Optionally, in this embodiment, the point cloud data of the target object is obtained by a scanning device on the target vehicle, wherein the scanning device may be, but is not limited to, a laser radar.

[0060] Optionally, in this embodiment, the point cloud data of the target object is mapped to a coordinate system, and the point cloud data of the target object is segmented based on a preset size or a preset step size to form equidistantly arranged columns as at least two original point cloud columns associated with the target object.

[0061] S202, determining target features of the point clouds included in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds included in each original point cloud column.

[0062] The center position refers to the coordinates of the center point of each original point cloud column. The target feature refers to the optimized features of each point cloud determined based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column.

[0063] Optionally, in this embodiment, the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column are input into the detection model, and the detection model outputs the target features of the point cloud contained in each original point cloud column. The detection model may be a neural network model.

[0064] S203: Detect the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0065] Optionally, in this embodiment, based on the target features of each original point cloud column and the point cloud contained in each original point cloud column, a pseudo image of the target object is determined. The pseudo image is input into the target detection model to obtain the target detection model to detect the target object and output the detection result. Specifically, in this embodiment, the target features of each original point cloud column and the point cloud contained in each original point cloud column are input into the SA module of the point cloud processing network (PointNet) to obtain a pseudo image of the target object. Among them, the SA module refers to the "Self-Attention" module. The target detection model refers to the model corresponding to the PointPillars algorithm. In this embodiment, the pseudo image is input into the target detection model to obtain the target detection model to detect the target object, and an optional implementation method of the output detection result is to input the pseudo image into the backbone network and detection head of the PointPillars algorithm in turn to detect the target object and output the detection result. The detection result can be the interval distance (the interval distance between the target vehicle and the target object), the height of the target object, etc.

[0066] The above-mentioned target detection method obtains at least two original point cloud columns associated with the target object by segmenting the point cloud data of the target object. Based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, the target features of the point cloud contained in each original point cloud column are determined. Based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected. Based on the target detection method of the present application, high-precision detection of the target object can be achieved, which effectively solves the problem of inaccurate detection results for smaller targets in the prior art.

[0067] In one embodiment, in order to more accurately determine the target features, such as Figure 3 As shown, an optional implementation of S202 includes:

[0068] S301, for each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column.

[0069] The original position refers to the position of the point cloud in the coordinate system. The relative position refers to the relative position of the point cloud contained in the original point cloud column and the center position of the original point cloud column to which it belongs.

[0070] Optionally, in this embodiment, for each original point cloud column, the geometric center of all point clouds included in the original point cloud column is determined according to the original positions of the point clouds included in the original point cloud column, and the center position corresponding to the geometric center is used as the center position of the original point cloud column.

[0071] S302, determining information features of the point cloud contained in the original point cloud column according to the center position, and the original position and relative position of the point cloud contained in the original point cloud column.

[0072] Optionally, in this embodiment, the information features of the point cloud contained in the original point cloud column are determined according to the center position, reflectivity information, and the original position and relative position of the point cloud contained in the original point cloud column. The reflectivity information refers to the reflectivity information when the laser radar scans the target object.

[0073] Optionally, the information feature of the point cloud in this embodiment can be represented as (x i ,y i ,z i ,r,x c ,y c ,z c ,x i-c ,y i-c ,z i-c ), where x i ,y i ,z iThe coordinates corresponding to the original position of the point cloud, r represents the reflectivity information, x c ,y c ,z c , is the coordinate corresponding to the center position of the original point cloud column to which the point cloud belongs, x i-c ,y i-c ,z i-c They are the relative positions of the original position and the center position of the point cloud respectively.

[0074] S303, based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

[0075] Optionally, in this embodiment, the information features of the point cloud contained in the original point cloud column are input into the attention model, and the attention model processes the information features of the point cloud contained in the original point cloud column to obtain the target features of the point cloud contained in the original point cloud column.

[0076] In this embodiment, for each original point cloud column, the relative position of the point cloud contained in the original point cloud column is determined based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column. According to the center position, and the original position and relative position of the point cloud contained in the original point cloud column, the information features of the point cloud contained in the original point cloud column are determined. Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed, and the target features of the point cloud contained in the original point cloud column can be determined more accurately.

[0077] In one embodiment, in order to further improve the accuracy of the target feature, the attention model includes a point attention model and a point column attention model, such as Figure 4 As shown, an optional implementation of S303 includes:

[0078] S401, inputting the information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column.

[0079] Among them, the point attention model refers to the attention model related to point cloud.

[0080] Optionally, in this embodiment, the information features of the point cloud contained in the original point cloud column are input into the maximum pooling layer to obtain the first descriptor output by the maximum pooling layer for pooling the information features of the point cloud. The information features of the point cloud contained in the original point cloud column are input into the average pooling layer to obtain the second descriptor output by the average pooling layer for pooling the information features of the point cloud. The first descriptor, the second descriptor and the information features are input into the fully connected layer to obtain the attention weighted features of the point cloud output by the fully connected layer for processing the first descriptor, the second descriptor and the information features.

[0081] In this embodiment, the first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature. An optional implementation method of the output point cloud attention weighted feature is to input the first descriptor and the second descriptor into the fully connected layer, add them after two layers of fully connected layers, and input the addition result into the Sigmoid function to obtain an attention feature map, multiply the attention feature map by the information feature, and obtain the attention weighted feature of the point cloud.

[0082] S402, using a point-column attention model, splicing the point-column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain splicing features of the point cloud.

[0083] Among them, the point column attention model refers to the attention model related to the point column. The point column feature refers to the global feature of the original point column.

[0084] Optionally, in this embodiment, the point column feature is input into the fully connected layer of the point column attention model, and after being mapped into a high-dimensional feature by the fully connected layer, it is concatenated with the attention weighted feature, and the concatenated feature is passed through two layers of fully connected layers in sequence to obtain the concatenated feature. Among them, the point column feature can be the arithmetic mean of each point cloud in the original point cloud column.

[0085] S403, performing a product operation on the attention weighted feature and the concatenated feature to obtain the target feature of the point cloud.

[0086] In this embodiment, the information features of the point cloud contained in the original point cloud column are input into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column. Using the point column attention model, the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column are spliced ​​to obtain the spliced ​​features of the point cloud. The attention weighted features and the spliced ​​features are multiplied to obtain the target features of the point cloud, which further increases the accuracy of the determined target features.

[0087] In one embodiment, if Figure 5 As shown, an optional implementation of a target detection method includes:

[0088] S501, segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object.

[0089] S502 : For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column.

[0090] S503: for each original point cloud column, determine the geometric center of all point clouds included in the original point cloud column according to the original position of each point cloud included in the original point cloud column.

[0091] S504: Taking the center position corresponding to the geometric center as the center position of the original point cloud column.

[0092] S505, determining information features of the point cloud included in the original point cloud column according to the center position, and the original position and relative position of the point cloud included in the original point cloud column.

[0093] S506, inputting the information features of the point cloud contained in the original point cloud column into the maximum pooling layer, obtaining a first descriptor outputted by the maximum pooling layer after performing pooling processing on the information features of the point cloud.

[0094] S507, inputting the information features of the point cloud contained in the original point cloud column into the average pooling layer, and obtaining a second descriptor outputted by the average pooling layer after performing pooling processing on the information features of the point cloud.

[0095] S508, input the first descriptor, the second descriptor and the information feature into the fully connected layer, obtain the attention weighted feature of the point cloud output by the fully connected layer after processing the first descriptor, the second descriptor and the information feature.

[0096] S509, using the point column attention model, splicing the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain the splicing features of the point cloud.

[0097] S510, performing a product operation on the attention weighted feature and the concatenated feature to obtain a target feature of the point cloud.

[0098] S511, determining a pseudo image of the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0099] S512, input the pseudo image into the target detection model, obtain the target detection model to detect the target object, and output the detection result.

[0100] In this embodiment, by segmenting the point cloud data of the target object, at least two original point cloud columns associated with the target object are obtained. Based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, the target features of the point cloud contained in each original point cloud column are determined. Based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected. Based on the target detection method of the present application, high-precision detection of the target object can be achieved, which effectively solves the problem of inaccurate detection results for smaller targets in the prior art.

[0101] Based on the same inventive concept, the embodiment of the present application also provides a target detection device for implementing the target detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the target detection device embodiment provided below can refer to the limitations of the target detection method above, and will not be repeated here.

[0102] In one embodiment, if Figure 6 As shown, a target detection device 1 is provided, comprising:

[0103] A segmentation module 10 is used to segment the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0104] A first determination module 20, for determining target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0105] The detection module 30 is used to detect the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0106] Based on the device in this embodiment, by segmenting the point cloud data of the target object, at least two original point cloud columns associated with the target object are obtained. Based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, the target features of the point cloud contained in each original point cloud column are determined. Based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected. Based on the target detection method of the present application, high-precision detection of the target object can be achieved, which effectively solves the problem of inaccurate detection results for smaller targets in the prior art.

[0107] In one embodiment, the Figure 6 The first determining module 10 is further specifically used for:

[0108] For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column;

[0109] According to the center position, and the original position and relative position of the point cloud contained in the original point cloud column, the information features of the point cloud contained in the original point cloud column are determined;

[0110] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

[0111] In one embodiment, the Figure 6The first determining module 10 is further specifically used for:

[0112] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column, including:

[0113] Input the information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column;

[0114] The point column attention model is used to splice the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain the splicing features of the point cloud;

[0115] The attention weighted features and the concatenated features are multiplied to obtain the target features of the point cloud.

[0116] In one embodiment, the column attention model includes a maximum pooling layer, an average pooling layer and a fully connected layer. Figure 6 The first determining module 10 is further specifically used for:

[0117] Input the information features of the point cloud contained in the original point cloud column into the maximum pooling layer, and obtain the first descriptor output by the maximum pooling layer after pooling the information features of the point cloud;

[0118] Input the information features of the point cloud contained in the original point cloud column into the average pooling layer, and obtain the second descriptor output by the average pooling layer after performing pooling processing on the information features of the point cloud;

[0119] The first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature, and outputs the attention weighted feature of the point cloud.

[0120] In one embodiment, the Figure 6 The detection module 30 is also specifically used for:

[0121] Determine a pseudo image of the target object based on each original point cloud column and target features of the point cloud contained in each original point cloud column;

[0122] The pseudo image is input into the target detection model, and the target detection model detects the target object and outputs the detection result.

[0123] In one embodiment, the Figure 6 The target detection device 1 further comprises:

[0124] A second determination module is used to determine, for each original point cloud column, the geometric center of all point clouds included in the original point cloud column according to the original position of each point cloud included in the original point cloud column;

[0125] The third determination module is used to use the center position corresponding to the geometric center as the center position of the original point cloud column.

[0126] Each module in the above target detection device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0127] In one embodiment, a computer device is provided, and its internal structure diagram can be shown in FIG7. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. 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 and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a target detection method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0128] Those skilled in the art will understand that the structure shown in FIG. 7 is merely a block diagram of a partial structure 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. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0129] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0130] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0131] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0132] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0133] In one embodiment, when the processor executes the computer program, the following steps are further implemented: based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, determining the target feature of the point cloud contained in each original point cloud column, including:

[0134] For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column;

[0135] According to the center position, and the original position and relative position of the point cloud contained in the original point cloud column, the information features of the point cloud contained in the original point cloud column are determined;

[0136] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

[0137] In one embodiment, when the processor executes the computer program, the following steps are further implemented: the attention model includes a point attention model and a point-column attention model;

[0138] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column, including:

[0139] Input the information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column;

[0140] The point column attention model is used to splice the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain the splicing features of the point cloud;

[0141] The attention weighted features and the concatenated features are multiplied to obtain the target features of the point cloud.

[0142] In one embodiment, when the processor executes the computer program, the following steps are also implemented: the point column attention model includes a maximum pooling layer, an average pooling layer, and a fully connected layer;

[0143] The information features of the point cloud contained in the original point cloud column are input into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column, including:

[0144] Input the information features of the point cloud contained in the original point cloud column into the maximum pooling layer, and obtain the first descriptor output by the maximum pooling layer after pooling the information features of the point cloud;

[0145] Input the information features of the point cloud contained in the original point cloud column into the average pooling layer, and obtain the second descriptor output by the average pooling layer after performing pooling processing on the information features of the point cloud;

[0146] The first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature, and outputs the attention weighted feature of the point cloud.

[0147] In one embodiment, when the processor executes the computer program, the following steps are further implemented: based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected, including:

[0148] Determine a pseudo image of the target object based on each original point cloud column and target features of the point cloud contained in each original point cloud column;

[0149] The pseudo image is input into the target detection model, and the target detection model detects the target object and outputs the detection result.

[0150] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0151] For each original point cloud column, determine the geometric center of all point clouds contained in the original point cloud column according to the original positions of the point clouds contained in the original point cloud column;

[0152] The center position corresponding to the geometric center is used as the center position of the original point cloud column.

[0153] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0154] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0155] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0156] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0157] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, determining the target feature of the point cloud contained in each original point cloud column, including:

[0158] For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column;

[0159] According to the center position, and the original position and relative position of the point cloud contained in the original point cloud column, the information features of the point cloud contained in the original point cloud column are determined;

[0160] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

[0161] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: the attention model includes a point attention model and a point-column attention model;

[0162] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column, including:

[0163] Input the information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column;

[0164] The point column attention model is used to splice the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain the splicing features of the point cloud;

[0165] The attention weighted features and the concatenated features are multiplied to obtain the target features of the point cloud.

[0166] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: the point column attention model includes a maximum pooling layer, an average pooling layer, and a fully connected layer;

[0167] The information features of the point cloud contained in the original point cloud column are input into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column, including:

[0168] Input the information features of the point cloud contained in the original point cloud column into the maximum pooling layer, and obtain the first descriptor output by the maximum pooling layer after pooling the information features of the point cloud;

[0169] Input the information features of the point cloud contained in the original point cloud column into the average pooling layer, and obtain the second descriptor output by the average pooling layer after performing pooling processing on the information features of the point cloud;

[0170] The first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature, and outputs the attention weighted feature of the point cloud.

[0171] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected, including:

[0172] Determine a pseudo image of the target object based on each original point cloud column and target features of the point cloud contained in each original point cloud column;

[0173] The pseudo image is input into the target detection model, and the target detection model detects the target object and outputs the detection result.

[0174] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0175] For each original point cloud column, determine the geometric center of all point clouds contained in the original point cloud column according to the original positions of the point clouds contained in the original point cloud column;

[0176] The center position corresponding to the geometric center is used as the center position of the original point cloud column.

[0177] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0178] Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object;

[0179] Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column;

[0180] The target object is detected based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

[0181] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: based on the center position of each original point cloud column and the original position of the point cloud contained in each original point cloud column, determining the target feature of the point cloud contained in each original point cloud column, including:

[0182] For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column;

[0183] According to the center position, and the original position and relative position of the point cloud contained in the original point cloud column, the information features of the point cloud contained in the original point cloud column are determined;

[0184] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

[0185] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: the attention model includes a point attention model and a point-column attention model;

[0186] Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column, including:

[0187] Input the information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column;

[0188] The point column attention model is used to splice the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column to obtain the splicing features of the point cloud;

[0189] The attention weighted features and the concatenated features are multiplied to obtain the target features of the point cloud.

[0190] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: the point column attention model includes a maximum pooling layer, an average pooling layer, and a fully connected layer;

[0191] The information features of the point cloud contained in the original point cloud column are input into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column, including:

[0192] Input the information features of the point cloud contained in the original point cloud column into the maximum pooling layer, and obtain the first descriptor output by the maximum pooling layer after pooling the information features of the point cloud;

[0193] Input the information features of the point cloud contained in the original point cloud column into the average pooling layer, and obtain the second descriptor output by the average pooling layer after performing pooling processing on the information features of the point cloud;

[0194] The first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature, and outputs the attention weighted feature of the point cloud.

[0195] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: based on each original point cloud column and the target features of the point cloud contained in each original point cloud column, the target object is detected, including:

[0196] Determine a pseudo image of the target object based on each original point cloud column and target features of the point cloud contained in each original point cloud column;

[0197] The pseudo image is input into the target detection model, and the target detection model detects the target object and outputs the detection result.

[0198] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0199] For each original point cloud column, determine the geometric center of all point clouds contained in the original point cloud column according to the original positions of the point clouds contained in the original point cloud column;

[0200] The center position corresponding to the geometric center is used as the center position of the original point cloud column.

[0201] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and 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 embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0202] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, 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, they should be considered to be within the scope of this specification.

[0203] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A target detection method, characterized in that: The method comprises: Segmenting the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object; Determine target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column; The target object is detected based on each original point cloud column and the target features of the point clouds contained in each original point cloud column.

2. The method according to claim 1, characterized in that The determining of target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column includes: For each original point cloud column, based on the center position of the original point cloud column and the original position of the point cloud contained in the original point cloud column, determine the relative position of the point cloud contained in the original point cloud column; Determining information features of the point cloud contained in the original point cloud column according to the center position, and the original position and relative position of the point cloud contained in the original point cloud column; Based on the attention model, the information features of the point cloud contained in the original point cloud column are processed to obtain the target features of the point cloud contained in the original point cloud column.

3. The method according to claim 2, characterized in that The attention model includes a point attention model and a point-column attention model; The information features of the point cloud contained in the original point cloud column are processed based on the attention model to obtain the target features of the point cloud contained in the original point cloud column, including: Inputting information features of the point cloud contained in the original point cloud column into the point attention model to determine attention weighted features of the point cloud contained in the original point cloud column; Using the point column attention model, the point column features of the original point cloud column and the attention weighted features of the point cloud contained in the original point cloud column are spliced ​​to obtain the splicing features of the point cloud; A product operation is performed on the attention weighted feature and the splicing feature to obtain a target feature of the point cloud.

4. The method according to claim 3, characterized in that The point column attention model includes a maximum pooling layer, an average pooling layer and a fully connected layer; The step of inputting information features of the point cloud contained in the original point cloud column into the point attention model to determine the attention weighted features of the point cloud contained in the original point cloud column includes: Inputting information features of the point cloud contained in the original point cloud column into the maximum pooling layer, and obtaining a first descriptor output by the maximum pooling layer after performing pooling processing on the information features of the point cloud; Inputting information features of the point cloud contained in the original point cloud column into the average pooling layer to obtain a second descriptor output by the average pooling layer after performing pooling processing on the information features of the point cloud; The first descriptor, the second descriptor and the information feature are input into the fully connected layer, and the fully connected layer processes the first descriptor, the second descriptor and the information feature, and outputs the attention weighted feature of the point cloud.

5. The method according to claim 1, characterized in that The detecting of the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column includes: Determining a pseudo image of the target object based on each original point cloud column and target features of the point cloud contained in each original point cloud column; The pseudo image is input into a target detection model to obtain a detection result outputted by the target detection model for detecting the target object.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: For each original point cloud column, according to the original position of each point cloud contained in the original point cloud column, determine the geometric center of all point clouds contained in the original point cloud column; The center position corresponding to the geometric center is used as the center position of the original point cloud column.

7. A target detection device, characterized in that: The device comprises: A segmentation module, used to segment the point cloud data of the target object to obtain at least two original point cloud columns associated with the target object; A first determination module, for determining target features of the point clouds contained in each original point cloud column based on the center position of each original point cloud column and the original position of the point clouds contained in each original point cloud column; The detection module is used to detect the target object based on each original point cloud column and the target features of the point cloud contained in each original point cloud column.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the target detection method according to any one of claims 1 to 6 are implemented.

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 steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images