Detection method and device of laser radar target frame, electronic equipment and storage medium

Abstract

The invention relates to a laser radar target frame detection method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the feature extraction according to point cloud information collected by a laser radar, generating a feature pseudo image, carrying out the target prediction through employing the feature pseudo image, and obtaining prediction result information, target point cloud information is determined based on the prediction result information and the point cloud information, the target point cloud information is point cloud information in the prediction target frame, point cloud size proportion information is determined according to the target point cloud information and the prediction result information, and the point cloud size proportion information is a size proportion between the prediction target frame and the target point cloud information. Then, target frame quality information corresponding to the predicted target frame is generated by adopting the point cloud size proportion information; the problem that the quality of the target frame cannot be evaluated in the automatic driving analysis process due to the uninterpretability of the output result of the deep learning algorithm is solved, and the quality of the target frame can be effectively evaluated in automatic driving.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, electronic equipment and storage medium for detecting a laser radar target frame. Background Art

[0002] With the development of society, the functions of intelligent equipment are constantly improving, and object detection is an indispensable link in many intelligent scenarios. Object detection refers to the technology of using computer technology to detect and identify objects in the surrounding environment. For example, in scenarios such as autonomous driving and robot control, it is necessary to identify objects in the surrounding environment, and the methods of object inspection can include point cloud data analysis, image recognition, infrared recognition, etc.; among them, point cloud data analysis is widely used in scenarios with high precision requirements, such as vehicle autonomous driving, due to its high accuracy.

[0003] The existing point cloud data analysis process for autonomous driving mainly relies on lidar to collect point cloud data, and uses a deep learning algorithm to perform object recognition processing on the point cloud data and output recognition results. The recognition results output by the deep learning algorithm are usually target boxes of the recognized objects, but the output results of the deep learning algorithm are unexplainable, that is, there is a problem that the quality of the target box cannot be assessed during the autonomous driving analysis process. Summary of the invention

[0004] The present application provides a method, device, electronic device and storage medium for detecting a laser radar target frame to solve the problem that the quality of the target frame cannot be assessed during the autonomous driving analysis process due to the unexplainable output results of a deep learning algorithm.

[0005] In a first aspect, the present application provides a method for detecting a laser radar target frame, comprising:

[0006] Feature extraction is performed based on the point cloud information collected by the laser radar to generate a feature pseudo image;

[0007] Using the characteristic pseudo image to perform target prediction, obtaining prediction result information, wherein the prediction result information includes a predicted target frame;

[0008] Based on the prediction result information and the point cloud information, determine target point cloud information, where the target point cloud information is the point cloud information within the predicted target box;

[0009] Determine point cloud size ratio information according to the target point cloud information and the prediction result information, wherein the point cloud size ratio information is the size ratio between the predicted target box and the target point cloud information;

[0010] The point cloud size ratio information is used to generate target frame quality information corresponding to the predicted target frame.

[0011] Optionally, determining point cloud size ratio information based on the target point cloud information and the prediction result information includes:

[0012] Extracting target frame size information corresponding to the predicted target frame from the prediction result information;

[0013] Generate target point cloud distribution size information based on the target point cloud information;

[0014] The target frame size information and the target point cloud distribution size information are used to perform size ratio calculation to generate the point cloud size ratio information.

[0015] Optionally, the target point cloud information includes at least two target point clouds, and generating target point cloud distribution size information according to the target point cloud information includes:

[0016] Determine a target coordinate system according to the predicted target frame;

[0017] Based on the target coordinate system, determining point cloud coordinate information corresponding to each target point cloud in the target point cloud information;

[0018] The point cloud coordinate information is used to generate the target point cloud distribution size information.

[0019] Optionally, the target coordinate system includes a horizontal coordinate axis, a longitudinal coordinate axis, and a vertical coordinate axis, and the point cloud coordinate information is used to generate the target point cloud distribution size information, including:

[0020] Based on the transverse coordinate axis, point cloud distribution calculation is performed in combination with the point cloud coordinate information to obtain a first transverse axis size, where the first transverse axis size is the distribution size of the target point cloud in the direction of the transverse coordinate axis;

[0021] According to the longitudinal coordinate axis, point cloud distribution calculation is performed in combination with the point cloud coordinate information to obtain a first longitudinal axis size, where the first longitudinal axis size is the distribution size of the target point cloud in the direction of the longitudinal coordinate axis;

[0022] Based on the vertical coordinate axis, point cloud distribution calculation is performed in combination with the point cloud coordinate information to obtain a first vertical axis size, where the first vertical axis size is a distribution size of the target point cloud in the direction of the vertical coordinate axis;

[0023] The target point cloud distribution size information is generated using the first horizontal axis size, the second horizontal axis size, and the third horizontal axis size.

[0024] Optionally, the using the target frame size information and the target point cloud distribution size information to perform size ratio calculation to generate the point cloud size ratio information includes:

[0025] Extracting a target frame edge size from the target frame size information;

[0026] Extracting the point cloud distribution edge size from the target point cloud distribution size information;

[0027] The size ratio of the point cloud is calculated by using the target frame edge size and the point cloud distribution edge size to generate the point cloud size ratio information.

[0028] Optionally, the size ratio calculation is performed using the target frame edge size and the point cloud distribution edge size to generate the point cloud size ratio information, including:

[0029] Extracting a first horizontal axis size, a first longitudinal axis size, and a first vertical axis size from the point cloud distribution edge size;

[0030] Extracting a second horizontal axis size, a second vertical axis size, and a second vertical axis size from the target frame edge size;

[0031] determining a first ratio between the first horizontal axis dimension and the second horizontal axis dimension, a second ratio between the first vertical axis dimension and the second vertical axis dimension, and a third ratio between the first vertical axis dimension and the second vertical axis dimension;

[0032] The point cloud size ratio information is generated using the first ratio, the second ratio and the third ratio.

[0033] Optionally, the predicted target frame includes a target frame surface, and the step of using the point cloud coordinate information to generate the target point cloud distribution size information includes:

[0034] Based on the target coordinate system, construct a point cloud distribution area corresponding to the surface of the target frame;

[0035] Using the point cloud coordinate information, determining the point cloud distribution surface size of the point cloud distribution area;

[0036] The target point cloud distribution size information is generated according to the point cloud distribution surface size.

[0037] Optionally, using the target frame size information and the target point cloud distribution size information to perform size ratio calculation to generate the point cloud size ratio information includes:

[0038] Extracting the target frame surface size from the target frame size information;

[0039] Extracting the point cloud distribution surface size from the target point cloud distribution size information;

[0040] The target frame surface size and the point cloud distribution surface size are used to perform size ratio calculation to generate the point cloud size ratio information.

[0041] Optionally, the using the point cloud size ratio information to generate target frame quality information corresponding to the predicted target frame includes:

[0042] Determine a preset size ratio interval corresponding to the point cloud size ratio information, and determine it as a target size interval;

[0043] Obtaining interval quality information corresponding to the target size interval;

[0044] The target frame quality information is generated using the interval quality information.

[0045] In a second aspect, the present application provides a laser radar target frame detection device, comprising:

[0046] A feature extraction module is used to extract features based on the point cloud information collected by the laser radar and generate a feature pseudo image;

[0047] A target prediction module, used to use the characteristic pseudo image to perform target prediction and obtain prediction result information, wherein the prediction result information includes a predicted target frame;

[0048] A first determination module, configured to determine target point cloud information based on the prediction result information and the point cloud information, wherein the target point cloud information is the point cloud information within the predicted target box;

[0049] A second determination module is used to determine point cloud size ratio information based on the target point cloud information and the prediction result information, wherein the point cloud size ratio information is a size ratio between the predicted target frame and the target point cloud information;

[0050] A generation module is used to use the point cloud size ratio information to generate target frame quality information corresponding to the predicted target frame.

[0051] In a third aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0052] Memory, used to store computer programs;

[0053] The processor is used to implement the laser radar target frame detection method described in any one of the first aspects when executing the program stored in the memory.

[0054] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for detecting a laser radar target frame as described in any one of the first aspects is implemented.

[0055] The embodiment of the present application performs feature extraction based on point cloud information collected by a laser radar to generate a feature pseudo image, uses the feature pseudo image to perform target prediction, and obtains prediction result information, wherein the prediction result information includes a predicted target box, and determines the target point cloud information based on the prediction result information and the point cloud information, wherein the target point cloud information is the point cloud information within the predicted target box, and determines the point cloud size ratio information based on the target point cloud information and the prediction result information, wherein the point cloud size ratio information is the size ratio between the predicted target box and the target point cloud information, and then uses the point cloud size ratio information to generate target box quality information corresponding to the predicted target box; thereby achieving the purpose of quality detection of the predicted target box, solving the problem in the prior related technology that the target box cannot be quality assessed during the autonomous driving analysis process due to the unexplainable output results of the deep learning algorithm, and being able to effectively determine the target box quality corresponding to the predicted target box, thereby being able to effectively assess the quality of the target box in autonomous driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a flow chart of a method for detecting a laser radar target frame provided in an embodiment of the present application;

[0057] Figure 2 A schematic diagram of a first horizontal axis dimension scene of a laser radar target frame detection method provided in an embodiment of the present application;

[0058] Figure 3 A schematic diagram of a first vertical axis dimension scene of a laser radar target frame detection method provided in an embodiment of the present application;

[0059] Figure 4 A schematic diagram of a first vertical axis dimension scene of a laser radar target frame detection method provided in an embodiment of the present application;

[0060] Figure 5 A schematic diagram of an application scenario of a laser radar target frame detection method provided in an embodiment of the present application;

[0061] Figure 6 A schematic diagram of another application scenario of a laser radar target frame detection method provided in an embodiment of the present application;

[0062] Figure 7 A schematic diagram of another application scenario of a laser radar target frame detection method provided in an embodiment of the present application;

[0063] Figure 8A schematic diagram of the structure of a laser radar target frame detection device provided in an embodiment of the present application;

[0064] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0066] In the existing related technologies, when using deep learning algorithms for object recognition, the target frame output by the deep learning algorithm may not match the actual scene due to the influence of algorithm parameters, operating status and other factors. For example, a target frame with a length of 2 meters is identified for an object with a length of 1 meter, which will cause a large error in the subsequent analysis and processing; and since the output content of the deep learning algorithm is unexplainable, there is a problem of being unable to determine the rationality of the target frame; and in the application process of autonomous driving, there is a high degree of reliance on deep learning algorithms, and the requirements for precision and accuracy in the specific implementation process of autonomous driving are also high. At this time, the unexplainability of the output results of the deep learning algorithm makes it impossible to perform quality assessment on the target frame during the autonomous driving process, that is, it is impossible to accurately determine the quality of the target frame, which will cause a decrease in the accuracy of autonomous driving control. Therefore, there is a problem of being unable to perform quality assessment on the target frame during the autonomous driving analysis process.

[0067] In order to solve the problem in the prior art that the quality of the target frame cannot be assessed during the autonomous driving analysis process due to the unexplainable output results of the deep learning algorithm, the present application provides a laser radar target frame detection method, device, electronic device and storage medium, by extracting features based on the point cloud information collected by the laser radar, generating a feature pseudo image, using the feature pseudo image to predict the target, and obtaining prediction result information, the prediction result information includes a predicted target frame, and determining the target point cloud information based on the prediction result information and the point cloud information, the target point cloud information is the point cloud information in the predicted target frame, and determining the point cloud size ratio information based on the target point cloud information and the prediction result information, the point cloud size ratio information is the size ratio between the predicted target frame and the target point cloud information, and then using the point cloud size ratio information to generate the target frame quality information corresponding to the predicted target frame; thereby achieving the purpose of quality detection of the predicted target frame, solving the problem in the prior art that the quality of the target frame cannot be assessed during the autonomous driving analysis process due to the unexplainable output results of the deep learning algorithm, and being able to effectively determine the target frame quality corresponding to the predicted target frame, and then being able to effectively assess the quality of the target frame in autonomous driving.

[0068] Figure 1 A flow chart of a method for detecting a laser radar target frame provided in an embodiment of the present application. This method can be applied to one or more electronic devices such as a vehicle (e.g., an autonomous driving vehicle), a client, a server, etc. In addition, the execution subject of this method can be hardware or software. When the above-mentioned execution subject is hardware, the execution subject can be one or more of the above-mentioned electronic devices. For example, a single electronic device can execute this method, or a plurality of electronic devices can cooperate with each other to execute this method. When the above-mentioned execution subject is software, this method can be implemented as multiple software or software modules, or as a single software or software module. It is not specifically limited here.

[0069] like Figure 1 As shown, a method for detecting a laser radar target frame provided in an embodiment of the present application may specifically include the following steps:

[0070] Step S110: extracting features based on the point cloud information collected by the laser radar to generate a feature pseudo image.

[0071] Specifically, the laser radar refers to a radar used to collect point cloud data, which can be installed on vehicles, robots and other subjects; and the point cloud information can represent part or all of the point cloud data collected by the laser radar, so that feature extraction can be performed on the point cloud information to generate a feature pseudo-image, wherein feature extraction can represent voxelization and voxel feature extraction of the point cloud information, and assigning values ​​to corresponding pixels based on the position of the voxels to form a feature pseudo-image.

[0072] It should be noted that the point cloud information may include information corresponding to the X-axis direction, Y-axis direction, Z-axis direction, and I intensity. The three-dimensional perception range for feature extraction in this embodiment can be 3 to 124.6 meters in the X-axis direction under the right-hand system, -36.8 to 36.8 meters in the Y-axis direction, and -2.5 meters to 4.5 meters in the Z-axis direction. The I intensity value range is 0-255, and the three-dimensional size of the voxel is (0.2 meters, 0.2 meters, 7 meters). The maximum number of point clouds in each voxel is 48; the above is only an example for explanation. The specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0073] Step S120: using the feature pseudo image to perform target prediction to obtain prediction result information, where the prediction result information includes a predicted target box.

[0074] Specifically, after obtaining the feature pseudo image, the feature pseudo image can be used to perform target prediction to obtain prediction result information, wherein the target prediction can use a pre-configured deep learning algorithm or neural network model to calculate and process the feature pseudo image, and generate prediction result information based on the calculation results output by the deep learning algorithm or neural network model, so that the prediction result information includes a predicted target box, and the predicted target box represents the target box generated after predicting the target, and the predicted target box can be used to identify the size and position of the recognized object, etc.; it should be noted that since the point cloud information can contain point cloud data of multiple objects, after target prediction, prediction result information containing one or more predicted target boxes can be obtained.

[0075] In one example, the feature pseudo image can be input into a perception network that is entirely a 2D convolutional neural network. After parsing the network prediction results, the prediction result information is obtained through low-confidence target filtering and non-maximum suppression operations. The prediction properties of the perception network can be similar to the Anchor-Free End-to-End FullyConvolutional Network (AFdetV2) network. For each target box, its classification score, intersection over Union (IOU) score, center point three-dimensional coordinates, centroid three-dimensional coordinates, three-dimensional size, heading angle attributes and other information will be predicted to generate prediction result information; of course, the above is only an example for illustration, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0076] Step S130: Determine target point cloud information based on the prediction result information and the point cloud information, where the target point cloud information is the point cloud information within the predicted target box.

[0077] Specifically, after obtaining the prediction result information, the target point cloud information can be determined based on the prediction result information and the point cloud information, wherein the target point cloud information is the point cloud information within the prediction target box. Since the prediction result information can contain one or more prediction target boxes, and different prediction target boxes can contain different point cloud information, the target point cloud information and the prediction target box can be in a one-to-one correspondence. In addition, in the process of specifically determining the target point cloud information, an algorithm or function can be used for calculation to obtain the target point cloud information corresponding to each prediction target box.

[0078] In one example, the prediction result information and point cloud information can be used as input to the Compute Unified Device Architecture (CUDA) core function for calculation to determine the point clouds belonging to each prediction target box, thereby obtaining the target point cloud information; of course, the above is only an example for illustration, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make any specific limitations on this.

[0079] Step S140: Determine point cloud size ratio information based on the target point cloud information and the prediction result information, where the point cloud size ratio information is the size ratio between the predicted target box and the target point cloud information.

[0080] Specifically, after determining the target point cloud information, the point cloud size ratio information can be determined based on the target point cloud information and the prediction result information. Since the prediction result information can include one or more prediction target boxes, and the target point cloud information is the point cloud information within the prediction target box, the size of the prediction target box and the point cloud distribution size of the target point cloud information can be used for calculation to generate the point cloud size ratio information, that is, the point cloud size ratio information can be the size ratio between the prediction target box and the target point cloud information; it should be noted that the point cloud size ratio information can include one or more size ratios, such as the size ratio in the length direction, the size ratio in the width direction, the size ratio in the height direction, the size ratio corresponding to the surface, etc., and this embodiment does not make specific limitations on this.

[0081] In this embodiment, the point cloud size ratio information can be used to clarify the size ratio between the current predicted target box and the actual point cloud information. The predicted target box is the target box predicted and output by the deep learning algorithm or neural network model, and the point cloud information is the real point cloud data actually collected. Therefore, the predicted target box can be intuitively judged whether it meets the requirements through the point cloud size ratio information. For example, in a smart driving vehicle scenario requiring higher precision, if the ratio represented by the point cloud size ratio information is too small, it means that the size of the predicted target box is much larger than the actual point cloud information size, and there is a problem of over-prediction. In this case, the predicted target box can be discarded or re-predicted, thereby achieving the goal of accurately and clearly interpreting and quantifying the target box predicted and output by the deep learning algorithm or neural network model, avoiding the problem of being unable to perform quality assessment of the target box during the autonomous driving analysis process due to the unexplainable output results of the deep learning algorithm, and being able to effectively determine the quality of the target box corresponding to the predicted target box, thereby being able to effectively assess the quality of the target box in autonomous driving.

[0082] In one example, the length of the predicted target box included in the prediction result information is 2 meters. At this time, based on the target point cloud information corresponding to the predicted target box, the distribution of each target point cloud in the predicted target box is determined, so that the point cloud distribution size in the length direction of the predicted target box can be calculated to be 1.5 meters. At this time, the size of the predicted target box of 2 meters and the point cloud distribution size of the target point cloud information of 1.5 meters can be used for calculation, and the resulting size ratio is 1.5:2. Then, the size ratio in other directions can be combined to generate point cloud size ratio information; of course, the above is only an example for illustration, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0083] Step S150: using the point cloud size ratio information to generate target frame quality information corresponding to the predicted target frame.

[0084] Specifically, after obtaining the point cloud size ratio information, the point cloud size ratio information can be used to generate target frame quality information corresponding to the predicted target frame. Since the accuracy standards for the predicted target frame are different in different application scenarios, the quality detection strategy corresponding to the application scenario can be obtained at this time. The quality detection strategy is used to perform quality detection on the point cloud size ratio information to obtain the target frame quality information corresponding to the generated predicted target frame.

[0085] In one example, when the size ratio corresponding to the point cloud size ratio information is 1.5:2, the quality detection strategy corresponding to the current application scenario is that the size ratio interval is 0-0.5, which belongs to low quality, 0.5-0.7 belongs to medium quality, and 0.7-1 belongs to high quality. At this time, it can be determined that the point cloud size ratio information input is of high quality, that is, the target frame quality information corresponding to the generated predicted target frame is high-quality target frame information; of course, the above is only an example for explanation, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0086] In this embodiment, feature extraction is performed based on the point cloud information collected by the laser radar to generate a feature pseudo image, and the feature pseudo image is used to predict the target to obtain prediction result information, wherein the prediction result information includes a predicted target frame, and the target point cloud information is determined based on the prediction result information and the point cloud information, and the target point cloud information is the point cloud information within the predicted target frame, and the point cloud size ratio information is determined based on the target point cloud information and the prediction result information, and the point cloud size ratio information is the size ratio between the predicted target frame and the target point cloud information, that is, the predicted target frame can be specifically associated with the real point cloud information through the point cloud size ratio information, which plays a role in quantifying and interpreting the prediction result information. The specific situation of the prediction result information can be expressed intuitively and clearly; then, the point cloud size ratio information is used to generate the target frame quality information corresponding to the predicted target frame; thereby achieving the purpose of quality detection of the predicted target frame, solving the problem in the existing related technology that the target frame cannot be evaluated in the process of autonomous driving analysis due to the unexplainability of the output results of the deep learning algorithm, and can effectively determine the quality of the target frame corresponding to the predicted target frame, and then effectively evaluate the quality of the target frame in autonomous driving.

[0087] In some optional embodiments of the present application, step S140 determines the point cloud size ratio information based on the target point cloud information and the prediction result information, and may specifically include the following sub-steps: extracting the target box size information corresponding to the predicted target box from the prediction result information; generating the target point cloud distribution size information based on the target point cloud information; and using the target box size information and the target point cloud distribution size information to perform size ratio calculation to generate point cloud size ratio information.

[0088] In this embodiment, after obtaining the target point cloud information and the prediction result information, the target frame size information corresponding to the predicted target frame can be extracted from the prediction result information. The target frame size information is used to represent the information corresponding to the size of the predicted target frame. The target frame size information can include the size information such as the length, width, height, and area of ​​each face of the predicted target frame; and the target point cloud distribution size information can also be generated based on the target point cloud information. The target point cloud distribution size information represents the size of the point cloud distribution in the predicted target frame. Since the target point cloud information represents the point cloud information in the predicted target frame, and each point cloud information has a different distribution position, the target point cloud distribution size information can include the length, width, height, and distribution area at each position of the point cloud distribution. Size information; and then the target frame size information and the target point cloud distribution size information can be used to perform size ratio calculation to generate point cloud size ratio information, wherein the specific size ratio calculation can be to extract the target size that matches each other from the target frame size information and the target point cloud distribution size information for ratio calculation, and obtain the point cloud size ratio information based on the calculated size ratio.

[0089] For example, the target frame size information includes the target frame length, target frame width, target frame height, and area size of the first side of the target frame; and the target point cloud distribution size information includes the point cloud distribution length, point cloud distribution width, and point cloud distribution height; at this time, the target frame length matches the point cloud distribution length, the target frame width matches the point cloud distribution width, and the target frame height matches the point cloud distribution height, that is, the size ratio between the target frame length and the point cloud distribution length, the size ratio between the target frame width and the point cloud distribution width, and the size ratio between the target frame height and the point cloud distribution height can be calculated, and the point cloud size ratio information is generated based on the calculated size ratio; of course, the above is only an example for illustration, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0090] It can be seen that in this embodiment, the target box size information corresponding to the predicted target box output by the model or algorithm is used to calculate the size ratio with the target point cloud distribution size information corresponding to the specific point cloud information actually collected, so as to generate point cloud size ratio information, so that the point cloud size ratio information can be used as the ratio between the real object object and the prediction result obtained after the model or algorithm is predicted, that is, the deviation between the output result of the model or algorithm and the real object can be intuitively quantified through the point cloud size ratio information. For example, the ratio represented by the point cloud size ratio information can be between 0-1. The larger the ratio represented by the point cloud size ratio information, the smaller the deviation between the output result of the model or algorithm and the real object, and the smaller the ratio represented by the point cloud size ratio information, the larger the deviation between the output result of the model or algorithm and the real object. The point cloud size ratio information can be used to quantify the predicted target box in the prediction result information to make it interpretable.

[0091] In some optional embodiments of the present application, the target point cloud information in the present embodiment may include at least two target point clouds, and the target point cloud distribution size information is generated based on the target point cloud information. Specifically, the following sub-steps may be included: determining the target coordinate system based on the predicted target frame; determining the point cloud coordinate information corresponding to each target point cloud in the target point cloud information based on the target coordinate system; and generating the target point cloud distribution size information using the point cloud coordinate information.

[0092] In this embodiment, in the process of generating the target point cloud distribution size information, the target coordinate system can be determined based on the predicted target frame. Since the prediction result information can contain one or more predicted target frames, in the process of determining the target coordinate system, a corresponding coordinate system can be established for each predicted target frame and the coordinate system can be determined as the target coordinate system. Alternatively, a coordinate system corresponding to the predicted result information can be established, which is applicable to the predicted target frame in each prediction result information, that is, each predicted target frame corresponds to the same target coordinate system. Of course, other methods of determining the target coordinate system can also be used, and this embodiment does not make specific limitations on this. After determining the target coordinate system, the point cloud coordinate information corresponding to each target point cloud in the target point cloud information can be determined based on the target coordinate system. The point cloud coordinate information can represent the coordinates of the target point cloud in the target coordinate system. Then, the point cloud coordinate information can be used to generate the target point cloud distribution size information, thereby determining the target point cloud distribution size information corresponding to the target point cloud information.

[0093] In some optional embodiments of the present application, the target coordinate system includes a horizontal coordinate axis, a longitudinal coordinate axis and a vertical coordinate axis, and the point cloud coordinate information is used to generate the target point cloud distribution size information, which may specifically include the following sub-steps: based on the horizontal coordinate axis, the point cloud coordinate information is combined to perform point cloud distribution calculation to obtain a first horizontal axis size, and the first horizontal axis size is the distribution size of the target point cloud in the direction of the horizontal coordinate axis; based on the longitudinal coordinate axis, the point cloud coordinate information is combined to perform point cloud distribution calculation to obtain a first longitudinal axis size, and the first longitudinal axis size is the distribution size of the target point cloud in the direction of the longitudinal coordinate axis; based on the vertical coordinate axis, the point cloud coordinate information is combined to perform point cloud distribution calculation to obtain a first vertical axis size, and the first vertical axis size is the distribution size of the target point cloud in the direction of the vertical coordinate axis; the first horizontal axis size, the first longitudinal axis size and the first vertical axis size are used to generate the target point cloud distribution size information.

[0094] In the present embodiment, in the process of using point cloud coordinate information to generate target point cloud distribution size information, distribution sizes in different directions can be generated according to different coordinate axes of the target coordinate system, that is, point cloud distribution calculation can be performed based on the horizontal coordinate axis in combination with the point cloud coordinate information to obtain a first horizontal axis size, and the first horizontal axis size is the distribution size of the target point cloud in the direction of the horizontal coordinate axis, wherein a specific point cloud distribution calculation method may be to determine the horizontal axis distance between the horizontal coordinates corresponding to every two point cloud coordinate information, and determine the maximum horizontal axis distance as the first horizontal axis size; point cloud distribution calculation can be performed based on the vertical coordinate axis in combination with the point cloud coordinate information to obtain a first vertical axis size, and the first vertical axis size is the distribution size of the target point cloud in the direction of the vertical coordinate axis, wherein a specific point cloud distribution calculation method may be to determine the vertical axis distance between the vertical coordinates corresponding to every two point cloud coordinate information, and determine the maximum vertical axis distance as the first vertical axis size; The point cloud coordinate information is used to perform point cloud distribution calculation to obtain a first vertical axis size, where the first vertical axis size is the distribution size of the target point cloud in the direction of the vertical coordinate axis. A specific point cloud distribution calculation method may be to determine the vertical axis distance between the vertical coordinates corresponding to every two point cloud coordinate information, and determine the maximum vertical axis distance as the first vertical axis size; then the first horizontal axis size, the first vertical axis size and the first vertical axis size may be used to generate target point cloud distribution size information. Specifically, the set of the first horizontal axis size, the first vertical axis size and the first vertical axis size may be determined as the target point cloud distribution size information, or the first horizontal axis size, the first vertical axis size and the first vertical axis size may be weightedly calculated to obtain the target point cloud distribution size information, or the target point cloud distribution size information may be generated based on the first horizontal axis size, the first vertical axis size and the first vertical axis size in combination with other size information. Of course, other methods may also be used, and this embodiment does not specifically limit this.

[0095] In one example, for each predicted target box, the target point cloud information corresponding to the predicted target box can be used to determine the distance between the frontmost coordinate position and the rearmost coordinate position of the target point cloud in the predicted target box in the horizontal axis direction as the first horizontal axis size, such as Figure 2 The distance between l_front and l_back shown in the figure; determine the distance between the frontmost coordinate position and the last coordinate position of the target point cloud in the predicted target frame in the longitudinal direction as the first longitudinal axis size, such as Figure 3 The distance between w_left and w_right shown in the figure; determine the distance between the frontmost coordinate position and the last coordinate position of the target point cloud in the predicted target frame in the vertical axis direction as the first vertical axis size, such as Figure 4 The distance between h_up and h_down shown in the figure; then the overall example of obtaining the first horizontal axis size, the first vertical axis size and the first vertical axis size can be referred to Figure 5 As shown in the figure, l is the first horizontal axis size, w is the first longitudinal axis size, and h is the first vertical axis size; of course, the above is only an example for explanation, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0096] In an optional embodiment of the present application, the target frame size information and the target point cloud distribution size information are used to perform size ratio calculation to generate point cloud size ratio information, which may specifically include the following sub-steps: extracting the target frame edge size from the target frame size information; extracting the point cloud distribution edge size from the target point cloud distribution size information; using the target frame edge size and the point cloud distribution edge size to perform size ratio calculation to generate point cloud size ratio information.

[0097] In this embodiment, in the process of generating point cloud size ratio information, the target box side size can be extracted from the target box size information, and the target box side size can represent the side size of the predicted target box, such as the horizontal axis size of the box, the vertical axis size of the box, or the vertical axis size of the box; and the point cloud distribution side size matching the target box side size can be extracted from the target point cloud distribution size information, and the point cloud distribution side size represents the distribution size of the target point cloud in the preset target box, such as the horizontal axis size of the distribution, the vertical axis size of the distribution, the vertical axis size of the distribution, etc.; in this embodiment, the extracted target box side size and the point cloud distribution side size can have a one-to-one correspondence, so that the corresponding target box side size and the point cloud distribution side size can be used for size ratio calculation, such as using the corresponding horizontal axis size of the box and the horizontal axis size of the distribution to calculate the size ratio. After all the corresponding target box side sizes and point cloud distribution side sizes are calculated for the size ratio, the calculated size ratio can be used to generate the point cloud size ratio information.

[0098] In an optional embodiment of the present application, the target frame edge size and the point cloud distribution edge size are used to perform size ratio calculation to generate point cloud size ratio information, which may specifically include the following sub-steps: extracting the first horizontal axis size, the first vertical axis size and the first vertical axis size from the point cloud distribution edge size; extracting the second horizontal axis size, the second vertical axis size and the second vertical axis size from the target frame edge size; determining a first ratio between the first horizontal axis size and the second horizontal axis size, a second ratio between the first vertical axis size and the second vertical axis size, and a third ratio between the first vertical axis size and the second vertical axis size; and using the first ratio, the second ratio and the third ratio to generate point cloud size ratio information.

[0099] In this embodiment, after obtaining the target frame edge size and the point cloud distribution edge size, the first horizontal axis size, the first vertical axis size and the first vertical axis size can be extracted from the point cloud distribution edge size, and the second horizontal axis size, the second vertical axis size and the second vertical axis size can be extracted from the target frame edge size, the second horizontal axis size represents the edge size of the predicted target frame in the horizontal axis of the target coordinate system, the second vertical axis size represents the edge size of the predicted target frame in the vertical axis of the target coordinate system, and the second vertical axis size represents the edge size of the predicted target frame in the vertical axis of the target coordinate system; thereby, the first ratio between the first horizontal axis size and the second horizontal axis size, the first vertical axis size and the second vertical axis size can be calculated. The second ratio, and the third ratio between the first vertical axis size and the second vertical axis size; then the first ratio, the second ratio and the third ratio can be used to generate point cloud size ratio information. The specific generation method can be to determine the set of the first ratio, the second ratio and the third ratio as the target point cloud distribution size information, or to perform weighted calculation on the first ratio, the second ratio and the third ratio to obtain the target point cloud distribution size information, or to generate the target point cloud distribution size information based on the first ratio, the second ratio and the third ratio in combination with other size information; of course, other methods can also be used, and this embodiment does not make specific limitations on this.

[0100] In an optional embodiment of the present application, the predicted target box includes the target box surface, and the point cloud coordinate information is used to generate the target point cloud distribution size information, which may specifically include the following sub-steps: based on the target coordinate system, constructing a point cloud distribution area parallel to the target box surface; using the point cloud coordinate information to determine the point cloud distribution surface size of the point cloud distribution area; based on the point cloud distribution surface size, generating the target point cloud distribution size information.

[0101] In this embodiment, in the process of generating target point cloud distribution size information, a point cloud distribution area corresponding to the target frame surface can be constructed based on the target coordinate system, wherein the point cloud distribution area represents the distribution area established close to the target frame surface, that is, one side of the point cloud distribution area coincides with the target frame surface, and the target frame surface represents the surface of a preset target frame, such as the target frame surface can include the front surface, the back surface, the left surface, the right surface, the upper surface, the lower surface, etc., and different point cloud distribution areas can be constructed for different target frame surfaces; then the point cloud coordinate information can be used to determine the point cloud distribution surface size within the point cloud distribution area, wherein the point cloud distribution surface size represents the distribution size of the point cloud distributed in the point cloud distribution area; then the point cloud distribution surface size can be used to generate the target point cloud distribution size information, specifically, the point cloud distribution surface size can be combined with the point cloud distribution edge size in the aforementioned embodiment to generate the target point cloud distribution size information.

[0102] In an optional embodiment of the present application, the target frame size information and the target point cloud distribution size information are used to perform size ratio calculation to generate point cloud size ratio information, which may specifically include the following sub-steps: extracting the target frame surface size from the target frame size information; extracting the point cloud distribution surface size from the target point cloud distribution size information; and using the target frame surface size and the point cloud distribution surface size to perform size ratio calculation to generate point cloud size ratio information.

[0103] Specifically, in the process of generating point cloud size ratio information, the target frame surface size can be extracted from the target frame size information. The target frame surface size can represent the size of the predicted target frame surface, specifically the size of the surface that fits the point cloud distribution area in the predicted target frame, and the point cloud distribution surface size is extracted from the target point cloud distribution size information; thereby, the target frame surface size and the point cloud distribution surface size can be used to perform size ratio calculation to generate point cloud size ratio information, wherein the size ratio calculation can be the ratio of the point cloud distribution surface size to the target frame surface size as the point cloud size ratio information.

[0104] In one example, if Figure 6-7 As shown in FIG. 1 , in the four directions of the front, back, left, and right of the predicted target frame, the point cloud distribution area corresponding to the target frame surface is constructed. The small frame corresponding to the lower side of each target frame surface can be constructed. The small frame is also the point cloud distribution area, as shown in FIG. Figure 6 The following is a small box corresponding to the left surface of the predicted target box. Figure 7The figure shows the small box corresponding to the front surface of the predicted target box; the specific construction size of the point cloud distribution area can be a preset construction size, and the preset construction size can be the predicted proportion corresponding to the target box surface, such as the height of the small box is 55% of the height of the predicted target box, and the width is 10% of the predicted target box; at this time, all point clouds belonging to the small box can be counted, and the distribution size of the point cloud in the small box in the direction perpendicular to the vertical axis and parallel to the target box surface can be calculated, that is, the point cloud distribution surface size, and the point cloud distribution surface size is used to calculate the ratio of the distribution and corresponding size of the front, back, left, right and lower half of the point cloud in the target box, where the corresponding size of the predicted target box is the size of the target box surface that the current small box fits. The ratio range is [0,1]. These attributes can effectively describe whether the point cloud is filled in a large amount near the corresponding surface area (in the small box in the schematic diagram). If a large amount of filling occurs, it can be proved that the observation features on this surface are relatively rich and the positioning on this surface is relatively accurate. Otherwise, it is considered that the surface prediction degree is high and there may be a certain position deviation.

[0105] In an optional embodiment of the present application, step S150 uses point cloud size ratio information to generate target frame quality information corresponding to the predicted target frame, which may specifically include the following sub-steps: determining a preset size ratio interval corresponding to the point cloud size ratio information, and determining it as the target size interval; obtaining interval quality information corresponding to the target size interval; and using the interval quality information to generate target frame quality information.

[0106] In this embodiment, in the process of generating target frame quality information, a preset size ratio interval corresponding to the point cloud size ratio information can be determined as the target size ratio interval, wherein the preset size ratio interval represents different pre-configured size ratio intervals, such as a 0-0.3 size ratio interval, a 0.3-0.5 size ratio interval, etc., and different point cloud size ratio information contains different point cloud size ratios. At this time, the preset size ratio interval to which the point cloud size ratio contained in the point cloud size ratio information belongs can be determined as the target size interval; then the interval quality information corresponding to the target size interval can be obtained, and different target size intervals can correspond to different pre-configured interval quality information, and the interval quality information represents the quality of the predicted target frame, such as the interval quality information can be quality information with low credibility or quality information with high credibility, etc., and then the interval quality information can be used to generate the target frame quality information, and the target frame quality information can include the interval quality information, and the target frame quality information represents the quality information corresponding to the predicted target frame for output or storage.

[0107] It can be seen that in this embodiment, the target frame quality information corresponding to the predicted target frame is generated by the point cloud size ratio information, so that the user can intuitively view the specific prediction instructions of the predicted target frame based on the target frame quality information, and can also clarify the specific basis corresponding to the current target frame quality information by viewing the point cloud size ratio information, which makes the predicted target frame explainable.

[0108] In addition, after generating the target frame quality information, the target frame quality information can also be assigned to its corresponding predicted target frame and output to the back-end algorithm as an attribute of the predicted target frame. The back-end algorithm can be a back-end tracking filtering algorithm, etc., so that the back-end algorithm can process the target frame in combination with the specific content of the target frame quality information during processing based on the predicted target frame; for example, in the case where high-feasibility quality information is required for autonomous driving, the predicted target frame corresponding to the target frame quality information with low credibility can be discarded; of course, the above is only an example for illustration, and the specific implementation example can be adaptively adjusted according to the scenario and needs, and this embodiment does not make specific limitations on this.

[0109] like Figure 8 As shown, the present application also discloses an embodiment, providing a laser radar target frame detection device, comprising:

[0110] The feature extraction module 810 is used to extract features based on the point cloud information collected by the laser radar and generate a feature pseudo image;

[0111] A target prediction module 820 is used to perform target prediction using the feature pseudo image to obtain prediction result information, wherein the prediction result information includes a predicted target frame;

[0112] A first determination module 830 is used to determine target point cloud information based on the prediction result information and the point cloud information, wherein the target point cloud information is the point cloud information within the predicted target box;

[0113] A second determination module 840 is used to determine point cloud size ratio information according to the target point cloud information and the prediction result information, wherein the point cloud size ratio information is a size ratio between the predicted target frame and the target point cloud information;

[0114] The generation module 850 is used to generate target frame quality information corresponding to the predicted target frame using the point cloud size ratio information.

[0115] In an optional implementation manner of the present application, the second determining module 840 may include:

[0116] An extraction unit, configured to extract target frame size information corresponding to the predicted target frame from the prediction result information;

[0117] A first generating unit, configured to generate target point cloud distribution size information according to the target point cloud information;

[0118] The second generating unit is used to calculate the size ratio by using the target frame size information and the target point cloud distribution size information to generate the point cloud size ratio information.

[0119] In an optional implementation manner of the present application, the target point cloud information includes at least two target point clouds, and the first generating unit may include:

[0120] A first determining subunit, configured to determine a target coordinate system according to the predicted target frame;

[0121] A second determining subunit is used to determine point cloud coordinate information corresponding to each of the target point clouds in the target point cloud information based on the target coordinate system;

[0122] The first generating subunit is used to generate the target point cloud distribution size information by using the point cloud coordinate information.

[0123] In an optional implementation manner of the present application, the target coordinate system includes a horizontal coordinate axis, a longitudinal coordinate axis and a vertical coordinate axis, and the first generating subunit may include:

[0124] A first point cloud distribution calculation subunit is used to perform point cloud distribution calculation based on the transverse coordinate axis and in combination with the point cloud coordinate information to obtain a first transverse axis size, where the first transverse axis size is a distribution size of the target point cloud in the direction of the transverse coordinate axis;

[0125] A second point cloud distribution calculation subunit is used to perform point cloud distribution calculation based on the longitudinal coordinate axis and in combination with the point cloud coordinate information to obtain a first longitudinal axis size, where the first longitudinal axis size is a distribution size of the target point cloud in the direction of the longitudinal coordinate axis;

[0126] A third point cloud distribution calculation subunit is used to perform point cloud distribution calculation based on the vertical coordinate axis and in combination with the point cloud coordinate information to obtain a first vertical axis size, where the first vertical axis size is a distribution size of the target point cloud in the direction of the vertical coordinate axis;

[0127] The third generating subunit is used to generate the target point cloud distribution size information by using the first horizontal axis size, the second horizontal axis size and the third horizontal axis size.

[0128] In an optional implementation manner of the present application, the second generating unit may include:

[0129] A first extraction subunit, configured to extract a target frame edge size from the target frame size information;

[0130] A second extraction subunit is used to extract the point cloud distribution edge size from the target point cloud distribution size information;

[0131] The first size ratio calculation subunit is used to perform size ratio calculation using the target frame edge size and the point cloud distribution edge size to generate the point cloud size ratio information.

[0132] In an optional implementation manner of the present application, the first size ratio calculation subunit may include:

[0133] A third extraction subunit is used to extract a first horizontal axis size, a first longitudinal axis size and a first vertical axis size from the point cloud distribution edge size;

[0134] A fourth extraction subunit, configured to extract a second horizontal axis size, a second vertical axis size, and a second vertical axis size from the target frame edge size;

[0135] a third determining subunit, configured to determine a first ratio between the first horizontal axis size and the second horizontal axis size, a second ratio between the first vertical axis size and the second vertical axis size, and a third ratio between the first vertical axis size and the second vertical axis size;

[0136] The fourth generating subunit is used to generate the point cloud size ratio information by using the first ratio, the second ratio and the third ratio.

[0137] In an optional implementation of the present application, the predicted target frame includes a target frame surface, and the first generating subunit may include:

[0138] A construction subunit, used to construct a point cloud distribution area corresponding to the surface of the target frame based on the target coordinate system;

[0139] A fourth determining subunit, configured to determine a point cloud distribution surface size of the point cloud distribution area using the point cloud coordinate information;

[0140] The fifth generating subunit is used to generate the target point cloud distribution size information according to the point cloud distribution surface size.

[0141] In an optional implementation manner of the present application, the second generating unit may include:

[0142] A fifth extraction subunit, configured to extract a target frame surface size from the target frame size information;

[0143] A sixth extraction subunit, configured to extract a point cloud distribution surface size from the target point cloud distribution size information;

[0144] The second size ratio calculation subunit is used to perform size ratio calculation using the target frame surface size and the point cloud distribution surface size to generate the point cloud size ratio information.

[0145] In an optional implementation of the present application, the generation module 850 may include:

[0146] A determination unit, configured to determine a preset size ratio interval corresponding to the point cloud size ratio information, and determine the preset size ratio interval as a target size ratio interval;

[0147] An acquisition unit, used to acquire interval quality information corresponding to the target size interval;

[0148] The third generating unit is used to generate the target frame quality information by using the interval quality information.

[0149] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.

[0150] like Fig. 9 As shown, the embodiment of the present application provides an air conditioner control device, including a processor 910, a communication interface 920, a memory 930 and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940.

[0151] Memory 930, used for storing computer programs;

[0152] In one embodiment of the present application, the processor 910 is used to execute the program stored in the memory 930 to implement the control method for detecting the laser radar target frame provided by any of the aforementioned method embodiments, by performing feature extraction based on the point cloud information collected by the laser radar, generating a feature pseudo image, using the feature pseudo image to predict the target, and obtaining prediction result information, the prediction result information includes a predicted target frame, and determining the target point cloud information based on the prediction result information and the point cloud information, the target point cloud information is the point cloud information within the predicted target frame, and determining the point cloud size ratio information based on the target point cloud information and the prediction result information, the point cloud size ratio information is the size between the predicted target frame and the target point cloud information Proportion, that is, the predicted target box can be specifically associated with the real point cloud information through the point cloud size ratio information, which plays a role in quantifying and interpreting the prediction result information, and can intuitively and clearly express the specific situation of the prediction result information; then the point cloud size ratio information is used to generate the target box quality information corresponding to the predicted target box; thereby achieving the purpose of quality detection of the predicted target box, solving the problem in the existing related technology that the target box cannot be evaluated in the process of autonomous driving analysis due to the unexplainable output results of the deep learning algorithm, and can effectively determine the quality of the target box corresponding to the predicted target box, and then effectively evaluate the quality of the target box in autonomous driving.

[0153] The embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the laser radar target frame detection method provided in any of the aforementioned method embodiments are implemented, by extracting features based on the point cloud information collected by the laser radar, generating a feature pseudo image, using the feature pseudo image to predict the target, and obtaining prediction result information, wherein the prediction result information includes a predicted target frame, and determining target point cloud information based on the prediction result information and the point cloud information, wherein the target point cloud information is the point cloud information within the predicted target frame, and determining point cloud size ratio information based on the target point cloud information and the prediction result information, wherein the point cloud size ratio information is the ratio between the predicted target frame and the target point cloud information. The size ratio between the information, that is, the predicted target box can be specifically associated with the real point cloud information through the point cloud size ratio information, which plays a role in quantifying and interpreting the prediction result information, and can intuitively and clearly express the specific situation of the prediction result information; then the point cloud size ratio information is used to generate the target box quality information corresponding to the predicted target box; thereby achieving the purpose of quality detection of the predicted target box, solving the problem in the existing related technology that the target box quality cannot be evaluated during the autonomous driving analysis process due to the unexplainable output results of the deep learning algorithm, and can effectively determine the target box quality corresponding to the predicted target box, and then effectively evaluate the quality of the target box in autonomous driving.

[0154] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0156] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0157] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for detecting a laser radar target frame, characterized in that: include: Feature extraction is performed based on the point cloud information collected by the laser radar to generate a feature pseudo image; Using the characteristic pseudo image to perform target prediction, obtaining prediction result information, wherein the prediction result information includes a predicted target frame; Based on the prediction result information and the point cloud information, determine target point cloud information, where the target point cloud information is the point cloud information within the predicted target box; Determine point cloud size ratio information according to the target point cloud information and the prediction result information, wherein the point cloud size ratio information is the size ratio between the predicted target box and the target point cloud information; The point cloud size ratio information is used to generate target frame quality information corresponding to the predicted target frame.

2. The method for detecting a laser radar target frame according to claim 1, characterized in that: Determining point cloud size ratio information based on the target point cloud information and the prediction result information includes: Extracting target frame size information corresponding to the predicted target frame from the prediction result information; Generate target point cloud distribution size information based on the target point cloud information; The target frame size information and the target point cloud distribution size information are used to perform size ratio calculation to generate the point cloud size ratio information.

3. The method for detecting a laser radar target frame according to claim 2, characterized in that: The target point cloud information includes at least two target point clouds, and generating target point cloud distribution size information according to the target point cloud information includes: Determine a target coordinate system according to the predicted target frame; Based on the target coordinate system, determining point cloud coordinate information corresponding to each target point cloud in the target point cloud information; The point cloud coordinate information is used to generate the target point cloud distribution size information.

4. The method for detecting a laser radar target frame according to claim 3, characterized in that: The target coordinate system includes a horizontal coordinate axis, a longitudinal coordinate axis and a vertical coordinate axis. The point cloud coordinate information is used to generate the target point cloud distribution size information, including: Based on the transverse coordinate axis, point cloud distribution calculation is performed in combination with the point cloud coordinate information to obtain a first transverse axis size, where the first transverse axis size is the distribution size of the target point cloud in the direction of the transverse coordinate axis; According to the longitudinal coordinate axis, point cloud distribution calculation is performed in combination with the point cloud coordinate information to obtain a first longitudinal axis size, where the first longitudinal axis size is the distribution size of the target point cloud in the direction of the longitudinal coordinate axis; Based on the vertical coordinate axis, point cloud distribution calculation is performed in combination with the point cloud coordinate information to obtain a first vertical axis size, where the first vertical axis size is a distribution size of the target point cloud in the direction of the vertical coordinate axis; The target point cloud distribution size information is generated using the first horizontal axis size, the first vertical axis size, and the first vertical axis size.

5. The method for detecting a laser radar target frame according to claim 2, characterized in that: The using the target frame size information and the target point cloud distribution size information to perform size ratio calculation to generate the point cloud size ratio information includes: Extracting a target frame edge size from the target frame size information; Extracting the point cloud distribution edge size from the target point cloud distribution size information; The size ratio of the point cloud is calculated by using the target frame edge size and the point cloud distribution edge size to generate the point cloud size ratio information.

6. The method for detecting a laser radar target frame according to claim 5, characterized in that: The size ratio calculation is performed using the target frame edge size and the point cloud distribution edge size to generate the point cloud size ratio information, including: Extracting a first horizontal axis size, a first longitudinal axis size, and a first vertical axis size from the point cloud distribution edge size; Extracting a second horizontal axis size, a second vertical axis size, and a second vertical axis size from the target frame edge size; determining a first ratio between the first horizontal axis dimension and the second horizontal axis dimension, a second ratio between the first vertical axis dimension and the second vertical axis dimension, and a third ratio between the first vertical axis dimension and the second vertical axis dimension; The point cloud size ratio information is generated using the first ratio, the second ratio and the third ratio.

7. The method for detecting a laser radar target frame according to claim 2, characterized in that: The predicted target frame includes a target frame surface, and the point cloud coordinate information is used to generate the target point cloud distribution size information, including: Based on the target coordinate system, construct a point cloud distribution area corresponding to the surface of the target frame; Using the point cloud coordinate information, determining the point cloud distribution surface size of the point cloud distribution area; The target point cloud distribution size information is generated according to the point cloud distribution surface size.

8. A laser radar target frame detection device, characterized in that: include: A feature extraction module is used to extract features based on the point cloud information collected by the laser radar and generate a feature pseudo image; A target prediction module, used to use the characteristic pseudo image to perform target prediction and obtain prediction result information, wherein the prediction result information includes a predicted target frame; A first determination module, configured to determine target point cloud information based on the prediction result information and the point cloud information, wherein the target point cloud information is the point cloud information within the predicted target box; A second determination module is used to determine point cloud size ratio information based on the target point cloud information and the prediction result information, wherein the point cloud size ratio information is a size ratio between the predicted target frame and the target point cloud information; A generation module is used to use the point cloud size ratio information to generate target frame quality information corresponding to the predicted target frame.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the laser radar target frame detection method described in any one of claims 1-7 when executing the program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting a laser radar target frame as described in any one of claims 1 to 7 is implemented.

Images