Method, device and electronic equipment for processing bounding box of object

Abstract

The present disclosure provides a bounding box processing method and device of an object and an electronic device, relates to the technical field of artificial intelligence, in particular to the technical field of computer vision, video processing, deep learning, and the like, and can be applied to scenarios such as automatic driving and smart city. The specific implementation scheme is as follows: obtaining a sample data set; identifying the sample data set to obtain bounding boxes of the same type of objects located in different detection regions; obtaining parameter values of all the bounding boxes in different detection regions respectively; determining an adjustment parameter based on the parameter values of all the bounding boxes in different detection regions; and adjusting initial bounding boxes of the same type of objects output by a detection network according to the adjustment parameter to obtain target bounding boxes.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to the fields of computer vision, video processing, and deep learning. Background Technology

[0002] Currently, when post-processing the detection bounding boxes of objects, the common approach is to randomly place objects of different categories into the current training point cloud scene with a certain probability. However, this method is only effective when the training data is small (e.g., 30,000 to 50,000). When the training data reaches 100,000 or more, this method will fail. Summary of the Invention

[0003] This disclosure provides a method, apparatus, and electronic device for processing object detection frames.

[0004] According to one aspect of this disclosure, a method for processing object detection boxes is provided, comprising: acquiring a sample dataset, wherein the sample dataset contains at least point cloud sample data from two different detection regions; identifying the sample dataset to obtain detection boxes of the same type of object located in different detection regions; acquiring parameter values ​​of all detection boxes in different detection regions respectively; determining adjustment parameters based on the parameter values ​​of all detection boxes in different detection regions; and adjusting the initial detection boxes of the same type of object output by the detection network according to the adjustment parameters to obtain target detection boxes, wherein the initial detection boxes are identified by the detection network from the point cloud data of the region to be detected in different detection regions.

[0005] According to another aspect of this disclosure, a method for processing detection boxes of another object is provided, comprising: acquiring point cloud data of a region to be detected; identifying an initial detection box from the point cloud data of the region to be detected based on a detection network; and adjusting the initial detection box into a target detection box according to adjustment parameters, wherein the adjustment parameters are determined based on parameter values ​​of all detection boxes in at least two different detection regions, and the detection boxes of the same type of object in different detection regions are obtained by identifying a sample dataset, the sample dataset containing at least point cloud sample data in different detection regions.

[0006] According to one aspect of this disclosure, an apparatus for processing object detection boxes is provided, comprising: a first acquisition unit for acquiring a sample dataset, wherein the sample dataset contains at least point cloud sample data from two different detection regions; a first recognition unit for recognizing the sample dataset to obtain detection boxes of the same type of object located in different detection regions; a second acquisition unit for acquiring parameter values ​​of all detection boxes in different detection regions respectively; a determination unit for determining adjustment parameters based on the parameter values ​​of all detection boxes in different detection regions; and a first adjustment unit for adjusting the initial detection boxes of the same type of object output by the detection network according to the adjustment parameters to obtain target detection boxes, wherein the initial detection boxes are identified by the detection network from the point cloud data of the region to be detected in different detection regions.

[0007] According to one aspect of this disclosure, a processing apparatus for detecting bounding boxes of another object is provided, comprising: a third acquisition unit for acquiring point cloud data of a region to be detected; a second recognition unit for recognizing an initial bounding box from the point cloud data of the region to be detected based on a detection network; and a second adjustment unit for adjusting the initial bounding box into a target bounding box according to adjustment parameters, wherein the adjustment parameters are determined based on parameter values ​​of all bounding boxes in at least two different detection regions, and the bounding boxes of the same type of object in different detection regions are obtained by recognizing a sample dataset, the sample dataset containing at least point cloud sample data from different detection regions.

[0008] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a processing method for an object detection frame according to an embodiment of this disclosure.

[0009] According to another aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute a method for processing a detection frame of an object according to embodiments of this disclosure.

[0010] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a method for processing a detection frame of an object according to embodiments of this disclosure.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 This is a flowchart of a method for processing an object detection frame according to an embodiment of the present disclosure;

[0014] Figure 2(a) is a flowchart of a method for processing a detection box of another object according to an embodiment of the present disclosure;

[0015] Figure 2(b) is a schematic diagram of a computer product for processing a detection frame of an object according to an embodiment of the present disclosure;

[0016] Figure 2(c) is a schematic diagram of the display interface of a computer product for performing a method for processing a detection box of an object according to an embodiment of the present disclosure;

[0017] Figure 3 This is a flowchart of a method for post-processing a lidar point cloud detection frame according to an embodiment of the present disclosure;

[0018] Figure 4 This is a schematic diagram of a processing apparatus for an object detection frame according to an embodiment of the present disclosure;

[0019] Figure 5 This is a schematic diagram of a processing apparatus for a detection frame of another object according to an embodiment of the present disclosure;

[0020] Figure 6 This is a schematic diagram of an electronic device used to implement the object detection frame processing method of the embodiments of this disclosure. Detailed Implementation

[0021] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0022] The following describes a method for processing an object detection box according to an embodiment of the present disclosure.

[0023] Figure 1 This is a flowchart of a method for processing an object detection box according to an embodiment of the present disclosure, such as... Figure 1 As shown, the method may include the following steps:

[0024] Step S102: Obtain the sample dataset.

[0025] In the technical solution provided by step S102 of this disclosure, the sample dataset contains point cloud sample data from at least two different detection areas.

[0026] In this embodiment, the aforementioned sample dataset is used to provide sample data for training the detection network and test sample data for testing the detection network. The sample dataset, referred to as the training set, includes multiple sets of point cloud data and may contain point cloud sample data from at least two different detection regions. For example, the point cloud sample data from the two different detection regions may be a source domain training set and a target domain training set. The source domain training set can be used to train the detection network and may be a training set for all cities. The target domain training set is used to test the detection network and may be a training set for the target city. Generally, the target domain training set contains fewer sample data and exhibits a significant domain gap compared to the sample data in the source domain training set. This domain gap can be used to represent the differences between the sample data in the target domain training set and the sample data in the source domain training set.

[0027] Optionally, the target city training set is used to test the detection network. It can be a sample dataset of a single city in the training set of all cities, or it can be a sample dataset of other cities. No specific limitation is made here.

[0028] Optionally, the point cloud sample data in two different detection regions can be obtained by autonomously dividing all the point cloud sample data in the sample dataset. For example, if the sample dataset contains 1 million sets of point cloud sample data, 800,000 sets of point cloud sample data can be divided as the source domain point cloud sample data and 200,000 sets of point cloud sample data can be divided as the target domain point cloud sample data. It should be noted that this is only an example and does not specifically limit the method of dividing the sample dataset into regions. Any method and process used to divide the sample dataset into regions is within the protection scope of the embodiments of this disclosure, and will not be described in detail here.

[0029] Step S104: Identify the sample dataset to obtain detection boxes of the same type of object located in different detection areas.

[0030] In the technical solution provided by step S104 of this disclosure, after obtaining the sample dataset, the sample dataset can be identified to obtain detection boxes of the same type of object located in different detection areas.

[0031] In this embodiment, point cloud sample data in the sample dataset can be identified to obtain detection boxes of the same type of object in different detection areas. The detection boxes can be used to represent the size of the same type of object located in different detection areas. The detection boxes can include three-dimensional detection boxes, such as detection boxes of the same type of object in all city training sets and detection boxes of the same type of object in the target city training set.

[0032] Optionally, the above-mentioned objects of the same type can be objects of the same category, such as small cars, large vehicles, or pedestrians. This is only an example and is not specifically limited.

[0033] Optionally, the method described in the embodiments of this disclosure can be applied to scenarios such as autonomous driving and smart cities. That is, the objects in the detection area can be objects that need to be detected in different scenarios, such as pedestrians and vehicles in autonomous driving scenarios, and streetlights and bridges in smart city scenarios. The type of detection object is not specifically limited here. Any type of object in the detection area is within the protection scope of the embodiments of this disclosure, and they are not listed here.

[0034] Step S106: Obtain the parameter values ​​of all detection boxes in different detection areas.

[0035] In the technical solution provided by step S106 of this disclosure, after identifying the sample dataset and obtaining the detection boxes of the same type of object located in different detection areas, the parameter values ​​of all detection boxes in different detection areas can be obtained respectively.

[0036] In this embodiment, after obtaining the detection boxes of the same type of object located in different detection regions, the parameter values ​​of all detection boxes in different detection regions can be statistically analyzed. For example, the parameter values ​​of all detection boxes in the target domain training set and the parameter values ​​of all detection boxes in the source domain training set. The parameter values ​​of the detection boxes can be the average length, width, and height of the detection boxes of the same type of object.

[0037] For example, for the small car category, the parameter values ​​of the small car detection boxes in the source domain training set may include: average length 4.5 meters, average width 1.8 meters, and average height 1.6 meters; the parameter values ​​of the small car detection boxes in the target domain training set may include: average length 3.8 meters, average width 1.6 meters, and average height 1.4 meters. This is only an example and does not limit the specific values ​​of the parameter values ​​of the detection boxes within the detection area.

[0038] Step S108: Determine the adjustment parameters based on the parameter values ​​of all detection frames in different detection areas.

[0039] In the technical solution provided by step S108 of this disclosure, after obtaining the parameter values ​​of all detection frames in different detection areas, the adjustment parameters can be determined based on the parameter values ​​of all detection frames in different detection areas.

[0040] In this embodiment, the parameter values ​​of all detection boxes within different detection regions can be compared and calculated to obtain adjustment parameters. These adjustment parameters can be ratios or differences between the parameter values ​​of all detection boxes within different detection regions, such as the ratio or difference between the parameter values ​​of all detection boxes in the target domain training set and the parameter values ​​of all detection boxes in the source domain training set. The adjustment parameters can include length adjustment parameters, width adjustment parameters, and height adjustment parameters. Specifically, a length adjustment parameter can be determined based on the length parameter values ​​of all detection boxes within different detection regions, and this length adjustment parameter is used to adjust the length of the initial detection boxes output by the detection network. A width adjustment parameter can be determined based on the width parameter values ​​of all detection boxes within different detection regions, and this width adjustment parameter is used to adjust the width of the initial detection boxes output by the detection network. A height adjustment parameter can be determined based on the height parameter values ​​of all detection boxes within different detection regions, and this height adjustment parameter is used to adjust the height of the initial detection boxes output by the detection network.

[0041] For example, for the large vehicle category, the parameter values ​​of the large vehicle detection boxes in the source domain training set can include: an average length of 8 meters, an average width of 1.9 meters, and an average height of 1.8 meters; the average length of the large vehicle detection boxes in the target domain training set is 7.5 meters, an average width of 1.8 meters, and an average height of 1.7 meters. Therefore, the length adjustment parameter can be 0.94, obtained by dividing the average length of the large vehicle detection boxes in the target domain training set by the average length of the large vehicle detection boxes in the source domain training set; the width adjustment parameter can be 0.95, obtained by dividing the average width of the large vehicle detection boxes in the target domain training set by the average width of the large vehicle detection boxes in the source domain training set; and the height adjustment parameter can be 0.94, obtained by dividing the average height of the large vehicle detection boxes in the target domain training set by the average height of the large vehicle detection boxes in the source domain training set.

[0042] For example, the length adjustment parameter can also be the difference between the average length of the large vehicle detection boxes in the target domain training set and the average length of the large vehicle detection boxes in the source domain training set, minus 0.5. The width adjustment parameter can also be the difference between the average width of the large vehicle detection boxes in the target domain training set and the average width of the large vehicle detection boxes in the source domain training set, minus 0.1. The height adjustment parameter can be the difference between the average height of the large vehicle detection boxes in the target domain training set and the average height of the large vehicle detection boxes in the source domain training set, minus 0.1.

[0043] It should be noted that the above content is only for illustrative purposes and does not specifically limit the calculation method of the adjustment parameters. Any method and process used to calculate the adjustment parameters are within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0044] Step S110: Adjust the initial detection boxes of the same type of objects output by the detection network according to the adjustment parameters to obtain the target detection boxes.

[0045] In the technical solution provided by step S110 of this disclosure, the initial detection boxes of the same type of object output by the detection network can be adjusted according to the adjustment parameters obtained in step S108 to obtain the target detection box. The detection network can be used to output the initial detection box, which can be the detection box identified by the detection network from the point cloud data of the area to be detected in different detection areas, such as the detection box of the vehicle to be detected or the detection box of the pedestrian to be detected. This is only an example and is not specifically limited.

[0046] Optionally, the size parameters of the target detection box can be the product of the parameters of the initial detection box and the adjustment parameters. That is, the length of the target detection box can be the product of the length of the initial detection box and the length adjustment parameter, the width of the target detection box can be the product of the width of the initial detection box and the width adjustment parameter, and the height of the target detection box can be the product of the height of the initial detection box and the height adjustment parameter. For example, if the length of the initial detection box is 3 meters, the width is 2 meters, and the height is 2 meters, and the length adjustment parameter is 0.94, the width adjustment parameter is 0.95, and the height adjustment parameter is 0.94, then the length of the target detection box is 2.82, the width of the target detection box is 1.9, and the height of the target detection box is 1.88.

[0047] Optionally, the size parameters of the target detection box can be the sum of the parameters of the initial detection box and the adjustment parameters. That is, the length of the target detection box can be the sum of the length of the initial detection box and the length adjustment parameter, the width of the target detection box can be the sum of the width of the initial detection box and the width adjustment parameter, and the height of the target detection box can be the sum of the height of the initial detection box and the height adjustment parameter. For example, if the length of the initial detection box is 3 meters, the width is 2 meters, and the height is 2 meters, and the length adjustment parameter is -0.5, the width adjustment parameter is -0.1, and the height adjustment parameter is -0.1, then the length of the target detection box is 2.9, the width of the target detection box is 1.9, and the height of the target detection box is 1.9.

[0048] Optionally, the point cloud data within the initial detection box can also be adjusted by adjusting the parameters. The point cloud data can be a set of vectors of the same type of object output by the detection network in a three-dimensional coordinate system. For example, if the coordinates of a point in the point cloud data are (2, 3, 4) and the length adjustment parameter is 2, the adjusted point cloud coordinates will be (4, 3, 4) when the length adjustment parameter is a ratio and (4, 3, 4) when the length adjustment parameter is a difference. This is just an example and is not a specific limitation.

[0049] Optionally, in this embodiment of the disclosure, non-maximum suppression (NMS) can also be performed on the adjusted detection box. NMS is a method for eliminating multiple duplicate detection boxes generated in the same space.

[0050] It should be noted that the above content is only for illustrative purposes and does not specifically limit the method of adjusting the initial detection box. Any method and process used to adjust the initial detection box is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0051] It should be further noted that in related technologies, the detection network is mainly tuned using the target domain training set, and then the final detection box of the object is output by the tuned detection network. Instead, this application determines the adjustment parameters by comparing the target domain training set with the source domain training set, and adjusts the initial detection box output by the detection network based on the adjustment parameters to obtain the final detection box. This achieves the purpose of smoothing out the gaps between different data domains, thereby effectively reducing the gaps between different data domains.

[0052] Through steps S102 to S110, a sample dataset is obtained; the sample dataset is identified to obtain detection boxes of the same type of object located in different detection regions; the parameter values ​​of all detection boxes in different detection regions are obtained respectively; based on the parameter values ​​of all detection boxes in different detection regions, adjustment parameters are determined; and the initial detection boxes of the same type of object output by the detection network are adjusted according to the adjustment parameters to obtain target detection boxes. In other words, this embodiment provides a statistical post-processing method for detection boxes. First, the parameter values ​​of all detection boxes in different detection regions of the sample dataset are calculated to obtain adjustment parameters. Then, the initial detection boxes output by the detection network are adjusted using the adjustment parameters to obtain target detection boxes. This achieves the goal of directly modifying the detection boxes through adjustment parameters, thereby smoothing out the gaps between different data domains. This solves the technical problem of not being able to effectively reduce the gaps between different data domains and achieves the technical effect of effectively reducing the gaps between different data domains.

[0053] The method described in this embodiment will now be described in further detail.

[0054] As an optional implementation, step S106, obtaining the parameter values ​​of all detection boxes in different detection areas, includes: obtaining the first parameter value of all detection boxes in the detection area; obtaining the second parameter value of all detection boxes in the area to be detected; and determining the adjustment parameter based on the first parameter value and the second parameter value.

[0055] In this embodiment, different detection regions may include detection regions. A detection region may be a source domain training set, such as a training set of all cities. By statistically analyzing the length, width, and height of all detection boxes within the detection region, the first parameter values ​​of all detection boxes within the detection region can be obtained. The region to be detected may be a target domain training set, such as a target city training set. By statistically analyzing the length, width, and height of all detection boxes within the region to be detected, the second parameter values ​​of all detection boxes within the region to be detected can be obtained. Then, by calculating the first parameter values ​​and the second parameter values, adjustment parameters are obtained to achieve the technical effect of saving computing resources and reducing the amount of computation.

[0056] Optionally, point cloud sample data within the detection area can be used to train a detection network, which can then be used to test within the detection area to output a detection bounding box for the object to be detected.

[0057] Optionally, the first parameter value may include the length, width, and height of the detection box of the object in the source domain training set. For example, the first parameter value of the detection box of vehicle A may be 5 meters in length, 2 meters in width, and 1.6 meters in height.

[0058] Optionally, the second parameter value may include the length, width, and height of the detection box of the object in the target domain training set. For example, the first parameter value of the detection box of vehicle B may be 4.8 meters in length, 1.8 meters in width, and 1.5 meters in height.

[0059] It should be noted that the above content is only for illustrative purposes and does not specifically limit the value of the first parameter and the second parameter. Any value used to represent the first parameter and the second parameter is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0060] As an optional implementation, obtaining the first parameter value of all detection boxes within the detection area includes: obtaining the first average parameter value of the parameter values ​​of all detection boxes within the detection area.

[0061] In this embodiment, the average value of the parameter values ​​included in the first parameter value of all detection boxes of the same object within the detection area can be calculated to obtain the first average parameter value, so as to achieve the technical effect of statistically analyzing all detection boxes of the same category of objects within the detection area. The first average parameter value may include the average length, average width, and average height of all detection boxes within the detection area.

[0062] For example, for a small car within the detection area, there are a total of three detection frames. The first parameter value of detection frame 1 can be a length of 3 meters, a width of 1.6 meters, and a height of 1.5 meters. The first parameter value of detection frame 2 can be a length of 3.1 meters, a width of 1.5 meters, and a height of 1.6 meters. The first parameter value of detection frame 3 can be a length of 2.9 meters, a width of 1.6 meters, and a height of 1.6 meters. Then the first average parameter value can be: an average length of 3 meters, an average width of 1.57 meters, and an average height of 1.57 meters.

[0063] It should be noted that the above content is only for illustrative purposes and does not specifically limit the calculation method of the first average parameter value. Any method and process used to calculate the first average parameter value is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0064] As an optional implementation, obtaining the second parameter value of all detection boxes within the detection area includes: obtaining the second average parameter value of the parameter values ​​of all detection boxes within the detection area.

[0065] In this embodiment, the average value of the parameter values ​​included in the second parameter value of all detection boxes of the same object in the detection area can be calculated to obtain the second average parameter value, so as to achieve the technical effect of statistically analyzing all detection boxes of the same type of object in the detection area. The second average parameter value may include the average length, average width, and average height of all detection boxes in the detection area.

[0066] For example, for a medium-sized vehicle in the area to be inspected, there are a total of three inspection boxes. The first parameter value of inspection box 1 can be a length of 4.1 meters, a width of 1.6 meters, and a height of 1.8 meters. The first parameter value of inspection box 2 can be a length of 4 meters, a width of 1.7 meters, and a height of 1.7 meters. The first parameter value of inspection box 3 can be a length of 4.2 meters, a width of 1.6 meters, and a height of 1.9 meters. Then the second average parameter value can be: an average length of 4.1 meters, an average width of 1.63 meters, and an average height of 1.8 meters.

[0067] It should be noted that the above content is only for illustrative purposes and does not specifically limit the calculation method of the second average parameter value. Any method and process used to calculate the second average parameter value is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0068] As an optional implementation, determining the adjustment parameter based on the first parameter value and the second parameter value includes: determining the adjustment parameter of the target parameter type based on the first parameter value of the target parameter type and the second parameter value of the target parameter type.

[0069] In this embodiment, the target parameter type may include a length parameter type, a width parameter type, and a height parameter type. The adjustment parameter of the target parameter type can be obtained by calculating the first parameter value and the second parameter value of the same target parameter type. The detection box is adjusted by adjusting the adjustment parameter to achieve the technical effect of effectively reducing the gap between different data domains. The adjustment parameter can be used to adjust the target size of the initial detection box, and the type of the target size can be the target parameter type.

[0070] Optionally, when the target parameter type is a length parameter type, the obtained adjustment parameter can be a length adjustment parameter, which can be used to adjust the length of the initial detection box; when the target parameter type is a width parameter type, the obtained adjustment parameter can be a width adjustment parameter, which can be used to adjust the width of the initial detection box; when the target parameter type is a height parameter type, the obtained adjustment parameter can be a height adjustment parameter, which can be used to adjust the height of the initial detection box. Here, the target parameter type is not specifically limited.

[0071] As an optional implementation method, determining the adjustment parameter based on the first parameter value and the second parameter value includes: comparing the second parameter value and the first parameter value to obtain a comparison result; and determining the adjustment parameter based on the comparison result.

[0072] In this embodiment, the comparison result between the second parameter value of all detection boxes of the same type of object in the detection area and the first parameter value of all detection boxes in the detection area can be determined as the adjustment parameter. The detection boxes can be directly adjusted by adjusting the parameter to achieve the technical effect of effectively reducing the gap between different data domains. The comparison result can be the ratio or difference between the second parameter value and the first parameter value.

[0073] Optionally, the adjustment parameters may include the ratio or difference between the average length value in the second parameter value and the average length value in the first parameter value, the ratio or difference between the average width value in the second parameter value and the average width value in the first parameter value, and the ratio or difference between the average height value in the second parameter value and the average height value in the first parameter value.

[0074] For example, if the average length in the second parameter value is 4 meters, the average width is 3 meters, and the average height is 2 meters, and the average length in the first parameter value is 5 meters, the average width is 4 meters, and the average height is 1.8 meters, then the adjustment parameters can include a ratio of 0.8 or a difference of -1 meter between the average length of 4 meters in the second parameter value and the average length of 5 meters in the first parameter value; a ratio of 0.75 or a difference of -1 meter between the average width of 3 meters in the second parameter value and the average width of 4 meters in the first parameter value; and a ratio of 1.9 or a difference of 0.2 meters between the average height of 2 meters in the second parameter value and the average height of 1.8 meters in the first parameter value.

[0075] It should be noted that the above content is only for illustrative purposes and does not specifically limit the calculation method of the adjustment parameter. Any method and process for determining the adjustment parameter based on the comparison result between the second parameter value and the first parameter value is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0076] As an optional implementation, step S110, adjusting the initial detection boxes of the same type of objects output by the detection network according to the adjustment parameters to obtain the target detection box includes: adjusting the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box.

[0077] In this embodiment, the target detection box may include the adjusted size of the initial detection box and / or the point cloud data within the adjusted initial detection box. The size of the initial detection box can be adjusted according to the adjustment parameters, and / or the point cloud data within the initial detection box can be adjusted according to the adjustment parameters to obtain the target detection box, thereby achieving the technical effect of effectively reducing the gap between different data domains.

[0078] Optionally, adjusting the size of the initial detection frame according to the adjustment parameters can be done by adjusting the length, width, and height of the initial detection frame. For example, if the initial detection frame has dimensions of 2 meters in length, 4 meters in width, and 6 meters in height, the adjustment parameters can include: length adjustment parameter 2, width adjustment parameter 3, and height adjustment parameter 4. When the adjustment parameters are ratios, the adjusted target detection frame has dimensions of 4 meters in length, 12 meters in width, and 24 meters in height; when the adjustment parameters are differences, the adjusted target detection frame has dimensions of 4 meters in length, 7 meters in width, and 10 meters in height. This is only an example and is not a specific limitation.

[0079] Optionally, adjusting the point cloud data within the initial detection box by adjusting parameters can be done by adjusting the point cloud coordinates. For example, if a point cloud coordinate is (2, 3, 4), the adjustment parameters can include: length adjustment parameter 2, width adjustment parameter 4, and height adjustment parameter 3. When the adjustment parameter is a ratio, the adjusted point cloud coordinates are (4, 12, 12); when the adjustment parameter is a difference, the adjusted point cloud coordinates are (4, 7, 7). This is just an example and is not a specific limitation.

[0080] It should be noted that the above content is only for illustrative purposes and does not specifically limit the method of adjusting the size of the initial detection box and / or the point cloud data within the initial detection box. Any method and process of adjusting the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0081] As an optional implementation, adjusting the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box includes: scaling the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box.

[0082] In this embodiment, the scaling operation can include zooming in and zooming out. The size of the initial detection box can be scaled according to the adjustment parameters, and the point cloud data within the initial detection box can be scaled according to the adjustment parameters to obtain the target detection box, thereby achieving the technical effect of effectively reducing the gap between different data domains.

[0083] Optionally, when the adjustment parameter is a ratio, if the adjustment parameter is a value greater than 1, the size of the initial detection box and the point cloud coordinates of the point cloud data within the initial detection box can be enlarged according to the adjustment parameter; if the adjustment parameter is a value less than 1, the size of the initial detection box and the point cloud coordinates of the point cloud data within the initial detection box can be reduced according to the adjustment parameter.

[0084] Optionally, when the adjustment parameter is a difference value, if the adjustment parameter is a value greater than 0, the size of the initial detection box and the point cloud coordinates of the point cloud data within the initial detection box can be enlarged according to the adjustment parameter; if the adjustment parameter is a value less than 0, the size of the initial detection box and the point cloud coordinates of the point cloud data within the initial detection box can be reduced according to the adjustment parameter.

[0085] It should be noted that the above content is only a preferred embodiment of scaling the size of the initial detection box and the point cloud data within the initial detection box according to the adjustment parameters. The specific operation method of scaling the size of the initial detection box and the point cloud data within the initial detection box according to the adjustment parameters is not limited here. All methods and processes of scaling the size of the initial detection box and the point cloud data within the initial detection box according to the adjustment parameters are within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0086] As an optional implementation, the method further includes: identifying point cloud data of the region to be detected based on the trained detection network to obtain an initial detection box; or, detecting point cloud data of the region to be detected based on the trained detection network to obtain an initial detection box, wherein the initial detection box can be used to adjust the parameters of the detection network.

[0087] In this embodiment, during network inference, the trained detection network can be used to identify the point cloud data of the region to be detected and output initial detection boxes. Here, network inference refers to using a pre-trained network that can predict normally to perform the detection task. Alternatively, during network training, the training detection network can be used to detect the point cloud data of the region to be detected and output initial detection boxes. Here, network training refers to comparing the initial detection boxes with the ground truth values ​​after the input data is passed through the detection network to obtain the loss, and then updating the detection network based on the loss to train the detection network to better perform the detection task and predict detection boxes.

[0088] The following describes another method for processing the detection box of an object according to an embodiment of this disclosure.

[0089] Figure 2(a) is a flowchart of another method for processing an object detection box according to an embodiment of the present disclosure. As shown in Figure 2(a), the method may include the following steps:

[0090] Step S202: Obtain point cloud data of the area to be detected.

[0091] In the technical solution provided by step S202 of this disclosure, the point cloud data can be a set of vectors of the object to be detected in a three-dimensional coordinate system. The point cloud data of the area to be detected can be obtained by LiDAR. The area to be detected can be the area to be detected in different scenarios. For example, in the autonomous driving scenario, in order to ensure the driving safety of the vehicle, the area to be detected can be the area near the vehicle; in the smart city scenario, the area to be detected can be a certain city area. This is only an example and is not specifically limited.

[0092] Step S204: Identify the initial detection box from the point cloud data of the area to be detected based on the detection network.

[0093] In the technical solution provided by step S204 of this disclosure, the detection network can be used to output the detection box of the object to be detected in the detection area. The initial detection box of the object to be detected can be identified from the point cloud data of the detection area through the detection network. The initial detection box can be the detection box obtained by performing preliminary detection on the object to be detected through the detection network.

[0094] Step S206: Adjust the initial detection box to the target detection box according to the adjustment parameters.

[0095] In the technical solution provided by step S206 of this disclosure, the size of the initial detection box and the point cloud coordinates of the point cloud data within the initial detection box can be adjusted according to the adjustment parameters to achieve the purpose of adjusting the initial detection box to the target detection box, thereby effectively reducing the gap between different data domains. The adjustment parameters are determined based on the parameter values ​​of all detection boxes in at least two different detection regions. The two different detection regions can be a source domain training set and a target domain training set. The source domain training set can be a training set of all cities, and the target domain training set can be a training set in the target city. The detection boxes of the same type of object in different detection regions can be obtained by recognizing the sample dataset, and the sample dataset contains at least point cloud sample data in different detection regions.

[0096] Optionally, the target detection box can be a detection box obtained by scaling the size of the initial detection box and the point cloud data within the initial detection box according to the adjustment parameters.

[0097] Figure 2(b) is a schematic diagram of a computer product for processing a detection box of an object according to an embodiment of the present disclosure. As shown in Figure 2(b), the above steps S202 to S206 can be executed by a computing device. The point cloud data of the area to be detected is input into the computing device. The computing device identifies the initial detection box from the point cloud data of the area to be detected based on the detection network. The initial detection box is adjusted into a target detection box according to the adjustment parameters. The adjusted target detection box is output on the display interface of the computing device to achieve the technical effect of effectively reducing the gap between different data domains.

[0098] Figure 2(c) is a schematic diagram of the display interface of a computer product for performing a method for processing object detection frames according to an embodiment of the present disclosure. As shown in Figure 2(c), the initial detection frame of the object to be detected output by the detection network and the target detection frame after adjusting the initial detection frame according to the adjustment parameters can be displayed on the display interface of the computer product. It should be noted that the display method of the computer product's display interface is not specifically limited here. Any display method of the display interface of the computer product for performing a method for processing object detection frames is within the protection scope of the present disclosure, and will not be described in detail here.

[0099] Through the above steps S202 to S206, point cloud data of the area to be detected is obtained; an initial detection box is identified from the point cloud data of the area to be detected based on the detection network; the initial detection box is adjusted into a target detection box according to the adjustment parameters, thereby solving the technical problem of not being able to effectively reduce the gap between different data domains and achieving the technical effect of effectively reducing the gap between different data domains.

[0100] The method for processing the detection box of the object in the present disclosure will be further described below with reference to preferred embodiments.

[0101] LiDAR plays a crucial role in autonomous driving systems. It can be used to create real-time 3D models of the vehicle's environment, especially in scenarios where the camera's visual perception system fails, such as in heavy fog or at night, thereby improving the safety of the autonomous driving system.

[0102] The radar point cloud 3D target detection task refers to the task of accurately perceiving the position, size, and orientation of 3D targets (such as vehicles and pedestrians) in the lidar point cloud coordinate system using lidar. At the same time, due to the great success of deep learning technology in the fields of computer vision and natural language processing, the radar point cloud 3D target detection task, as a classic subtask in computer vision, has also become a hot topic for deep learning researchers.

[0103] In radar point cloud 3D target detection tasks, the data collected by lidar is generally displayed and processed in the form of point clouds. A point cloud can be understood as N points scattered in space, each point containing three floating-point values ​​(X, Y, Z) to represent its spatial location and one value (R) to represent its echo intensity. However, due to the irregularity, disorder, and sparsity of point cloud data collected by lidar, there are significant domain gaps between different batches of collected and labeled data. This leads to inconsistencies in the collected data. Furthermore, factors such as inconsistent hardware, different cities where data is collected, imbalanced data categories, and inconsistent labeling standards among multiple labelers can all have a significant negative impact on the performance of radar point cloud 3D target detection tasks.

[0104] One related technique involves training a teacher model using training data from a large number of different cities, then collecting a large amount of unlabeled data in the target city, using the teacher model to generate pseudo-labels on the unlabeled data, and finally training a student model on the pseudo-labeled dataset of the target city to solve the domain gap problem. However, training a large model and generating pseudo-labels requires a huge amount of data and computation, resulting in high costs.

[0105] In another related technique, by increasing the target domain data to smooth out the domain gaps, the data collected from the target city is converted into a point cloud object database. Then, during training, different categories of objects are randomly placed into the current training point cloud scene with a certain probability. However, this method is only effective for small data (e.g., 30,000 to 50,000). When the training data reaches 100,000 or more, this method cannot be used.

[0106] However, this disclosure provides a statistical post-processing method for lidar point cloud detection boxes. By statistically analyzing the data used during lidar point cloud training and performing large-scale statistical analysis on the overall data, the predicted boxes can be directly modified in the post-processing stage of the trained point cloud detection model to reduce the gap between the two data domains, thereby improving the point cloud detection effect.

[0107] Figure 3 This is a flowchart of a post-processing method for lidar point cloud detection boxes according to an embodiment of the present disclosure, such as... Figure 3 As shown, the method may include the following steps:

[0108] Step S301: Obtain statistical information for all city training sets and the target city training set.

[0109] In step S301 of this embodiment, the statistical information can be the average length, width and height of the detection box. The average length, width and height of the detection box refers to the average length, width and height of objects of the same type. For example, for the car category, the average length is 4.5 meters, and for the large car category, the average length is 8 meters. The statistics can be performed on all city training sets to obtain the average length, width and height of the detection boxes in all city training sets. The statistics can also be performed on the target city training set to obtain the average length, width and height of the detection boxes in the target city training set.

[0110] Optionally, the training set is obtained by dividing all collected data. For example, if a total of 1 million data points are collected, 800,000 data points are divided into a training set and 200,000 data points are divided into a test set. The training set mainly contains 1 million frames of point cloud data and ground truth detection boxes of objects contained in each frame of point cloud.

[0111] Optionally, the target city training set contains relatively little data and has significant domain gaps compared to the data in the training sets of all cities. Typically, when an algorithm is trained on the training sets of all cities and tested on the target city training set, its performance will drop significantly and it will not achieve the same performance as the training sets of all cities.

[0112] Step S302: Input the detection boxes output by the detection network and the statistical information of all city training sets and the target city training set into the calculation unit.

[0113] In step S302 of the present disclosure embodiment, during network training or network inference, the detection boxes output by the detection network and the statistical information of the two datasets can be input into the computing unit to calculate the adjustment coefficient of the currently output detection box, adjust the predicted detection box, and obtain the adjusted detection box.

[0114] Alternatively, training the network means that after the input data is passed through the network to obtain a prediction, it is compared with the ground truth to obtain a loss. Then the network updates the network based on the loss so that it can learn how to better predict detection boxes. Network inference usually refers to using a pre-trained network that can make predictions normally to perform detection tasks, but without calculating the loss or updating the network.

[0115] Optionally, the adjustment coefficient can be the average length of the detection boxes in the target city training set divided by the average length of the detection boxes in all city training sets, the average width of the detection boxes in the target city training set divided by the average width of the detection boxes in all city training sets, and the average height of the detection boxes in the target city training set divided by the average height of the detection boxes in all city training sets. This is only an example and is not specifically limited.

[0116] Optionally, adjusting the predicted detection box by adjusting the adjustment coefficient can be done by adjusting the length, width, and height of the detection box output by the detection network to the product of the length, width, and height of the detection box output by the detection network and the adjustment coefficient. For example, if the length adjustment coefficient of the computation unit output is 1.05 and the length of the detection box output by the network is 4 meters, then the adjusted length is 4.2 meters. At the same time, the point cloud within the detection box will also be scaled on the x-axis by 1.05. It should be noted that this is only an example of length using the adjustment coefficient, and no specific limitation is made on the adjustment coefficient. All methods and processes for adjusting the predicted detection box by adjusting the adjustment coefficient are within the protection scope of this disclosure.

[0117] Step S303: Perform non-maximum suppression on the adjusted detection box and output the final result.

[0118] In step S303 of the present embodiment, the final result can be all predicted boxes of the current frame, for example (bbox: x, y, z, h, w, l, yaw). NMS operation can be performed on the detection boxes adjusted according to the adjustment coefficient, and the predicted boxes after the operation can be output. NMS is a method for eliminating multiple duplicate detection boxes generated in the same space.

[0119] In steps S301 to S303 of this embodiment, statistical information is obtained by statistically analyzing the data of all city training sets and the target city training set. Then, the detection boxes output by the detection network and the statistical information of all city training sets and the target city training set are input into the calculation unit to obtain adjustment coefficients. The predicted detection boxes are directly modified according to the adjustment coefficients to achieve the goal of not increasing the amount of training or inference computation, and still smoothing out the domain gaps between data under large training data, thereby achieving the technical effect of steadily improving the 3D detection performance.

[0120] This disclosure also provides an embodiment for performing Figure 1 The illustrated embodiment describes a method for processing the detection box of an object and an apparatus for processing the detection box of an object.

[0121] Figure 4 This is a schematic diagram of a processing apparatus for an object detection frame according to an embodiment of the present disclosure, as shown below. Figure 4 As shown, the processing device 400 for the detection frame of the object may include: a first acquisition unit 401, a first identification unit 402, a second acquisition unit 403, a determination unit 404, and a first adjustment unit 405.

[0122] The first acquisition unit 401 is used to acquire a sample dataset, wherein the sample dataset contains point cloud sample data from at least two different detection regions.

[0123] The first recognition unit 402 is used to recognize the sample dataset and obtain detection boxes of the same type of objects located in different detection areas.

[0124] The second acquisition unit 403 is used to acquire the parameter values ​​of all detection boxes in different detection areas respectively.

[0125] The determining unit 404 is used to determine the adjustment parameters based on the parameter values ​​of all detection frames in different detection areas.

[0126] The first adjustment unit 405 is used to adjust the initial detection boxes of the same type of objects output by the detection network according to the adjustment parameters to obtain the target detection boxes, wherein the initial detection boxes are identified by the detection network from the point cloud data of the areas to be detected in different detection regions.

[0127] Optionally, the second acquisition unit 403 includes: a first acquisition module, used to acquire first parameter values ​​of all detection boxes within the detection area; a second acquisition module, used to acquire second parameter values ​​of all detection boxes within the area to be detected; and a determination module, used to determine adjustment parameters based on the first parameter values ​​and the second parameter values.

[0128] Optionally, the first acquisition module includes: a first acquisition submodule, used to acquire the first average parameter value of the parameter values ​​of all detection boxes within the detection area.

[0129] Optionally, the second acquisition module includes: a second acquisition submodule, used to acquire the second average parameter value of the parameter values ​​of all detection boxes within the detection area.

[0130] Optionally, the determining module includes: a first determining submodule, used to determine an adjustment parameter of the target parameter type based on a first parameter value and a second parameter value of the target parameter type, wherein the adjustment parameter is used to adjust the target size of the initial detection box, and the type of the target size is the target parameter type.

[0131] Optionally, the determining module includes: a comparison submodule, used to compare the second parameter value and the first parameter value to obtain a comparison result; and a second determining submodule, used to determine the adjustment parameter based on the comparison result.

[0132] Optionally, the first adjustment unit 405 includes: an adjustment module, used to adjust the size of the initial detection box and / or the point cloud data within the initial detection box according to adjustment parameters to obtain a target detection box, wherein the target detection box includes the adjusted size of the initial detection box and / or the point cloud data within the adjusted initial detection box.

[0133] Optionally, the adjustment module includes a scaling operation submodule, used to scale the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box.

[0134] Optionally, the device further includes: a third recognition unit, used to recognize the point cloud data of the region to be detected based on the trained detection network to obtain an initial detection box; or, a detection unit, used to detect the point cloud data of the region to be detected based on the trained detection network to obtain an initial detection box, wherein the initial detection box is used to adjust the parameters of the detection network.

[0135] In the object detection bounding box processing apparatus of this disclosure embodiment, a first acquisition unit is used to acquire a sample dataset, wherein the sample dataset contains at least point cloud sample data from two different detection regions; a first identification unit is used to identify the sample dataset to obtain detection bounding boxes of the same type of object located in different detection regions; a second acquisition unit is used to acquire parameter values ​​of all detection bounding boxes in different detection regions respectively; a determination unit is used to determine adjustment parameters based on the parameter values ​​of all detection bounding boxes in different detection regions; and a first adjustment unit is used to adjust the initial detection bounding boxes of the same type of object output by the detection network according to the adjustment parameters to obtain target detection bounding boxes, wherein the initial detection bounding boxes are identified by the detection network from the point cloud data of the region to be detected in different detection regions, thereby solving the technical problem of not being able to effectively reduce the gap between different data domains and achieving the technical effect of effectively reducing the gap between different data domains.

[0136] This disclosure also provides an apparatus for processing object detection frames for performing the object detection frame processing method of the embodiment shown in FIG2.

[0137] Figure 5 This is a schematic diagram of a processing apparatus for a detection frame of another object according to an embodiment of the present disclosure, such as... Figure 5 As shown, the processing device 500 for the detection frame of the object may include: a third acquisition unit 501, a second identification unit 502, and a second adjustment unit 503.

[0138] The third acquisition unit 501 is used to acquire point cloud data of the area to be detected.

[0139] The second recognition unit 502 is used to identify the initial detection box from the point cloud data of the area to be detected based on the detection network.

[0140] The second adjustment unit 503 is used to adjust the initial detection box to the target detection box according to the adjustment parameters. The adjustment parameters are determined based on the parameter values ​​of all detection boxes in at least two different detection regions. The detection boxes of the same type of object in different detection regions are obtained by identifying the sample dataset, and the sample dataset contains at least point cloud sample data in different detection regions.

[0141] In the object detection box processing apparatus of this disclosure embodiment, a third acquisition unit is used to acquire point cloud data of the region to be detected; a second identification unit is used to identify an initial detection box from the point cloud data of the region to be detected based on a detection network; and a second adjustment unit is used to adjust the initial detection box into a target detection box according to adjustment parameters. The adjustment parameters are determined based on the parameter values ​​of all detection boxes in at least two different detection regions. The detection boxes of the same type of object in different detection regions are obtained by identifying a sample dataset. The sample dataset contains at least point cloud sample data from different detection regions, thereby solving the technical problem of not being able to effectively reduce the gap between different data domains and achieving the technical effect of effectively reducing the gap between different data domains.

[0142] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0143] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0144] Figure 6 This is a schematic diagram of an electronic device used to implement a method for processing a detection frame of an object according to embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0145] like Figure 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0146] Multiple components in device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0147] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as the object detection frame processing method. For example, in some embodiments, the object detection frame processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the object detection frame processing method described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform the object detection frame processing method by any other suitable means (e.g., by means of firmware).

[0148] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0149] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0150] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0151] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0152] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0153] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0154] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0155] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for processing a detection bounding box of an object, comprising: Obtain a sample dataset, wherein the sample dataset contains point cloud sample data from at least two different detection regions, wherein the two different detection regions include a region to be detected and a detection region, the region to be detected is a target domain training set, the detection region is a source domain training set, and there is a domain gap between the sample data in the target domain training set and the sample data in the source domain training set, the domain gap being used to characterize the differences between the sample data in the target domain training set and the sample data in the source domain training set; The sample dataset is identified to obtain detection boxes of the same type of object located in the different detection regions; Obtain the parameter values ​​of all detection boxes within the different detection areas, wherein the objects corresponding to all detection boxes are of the same type. Based on the parameter values ​​of all detection frames within the different detection areas, the adjustment parameters are determined; The initial detection boxes of the same type of object output by the detection network are adjusted according to the adjustment parameters to obtain the target detection box, wherein the initial detection box is identified by the detection network from the point cloud data of the region to be detected in the different detection regions; The point cloud sample data within the detection area is used to train the detection network. Obtaining the parameter values ​​of all detection boxes within different detection areas includes: obtaining the first parameter value of all detection boxes within the detection area; obtaining the second parameter value of all detection boxes within the area to be detected; and determining the adjustment parameter based on the first parameter value and the second parameter value. Determining the adjustment parameter based on the first parameter value and the second parameter value includes: comparing the second parameter value and the first parameter value to obtain a comparison result; and determining the adjustment parameter based on the comparison result.

2. The method according to claim 1, wherein, Obtaining the first parameter values ​​of all detection boxes within the detection area includes: Obtain the first average parameter value of all detection boxes within the detection area.

3. The method according to claim 1, wherein obtaining the second parameter values ​​of all detection frames within the detection area comprises: Obtain the second average parameter value of the parameter values ​​of all detection boxes within the area to be detected.

4. The method according to claim 1, wherein determining the adjustment parameter based on the first parameter value and the second parameter value comprises: Based on the first parameter value and the second parameter value of the target parameter type, the adjustment parameter of the target parameter type is determined, wherein the adjustment parameter is used to adjust the target size of the initial detection box, and the type of the target size is the target parameter type.

5. The method according to claim 1, wherein adjusting the initial detection boxes of the same type of objects output by the detection network according to the adjustment parameters to obtain target detection boxes includes: The size of the initial detection box and / or the point cloud data within the initial detection box are adjusted according to the adjustment parameters to obtain the target detection box, wherein the target detection box includes the adjusted size of the initial detection box and / or the adjusted point cloud data within the initial detection box.

6. The method according to claim 5, wherein adjusting the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box comprises: The target detection box is obtained by scaling the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters.

7. The method according to any one of claims 1 to 6, further comprising: The initial detection box is obtained by identifying the point cloud data of the region to be detected based on the trained detection network. or, The detection network trained on the test network is used to detect point cloud data of the region to be detected, and the initial detection box is obtained. The initial detection box is used to adjust the parameters of the detection network.

8. A method for processing a detection box of an object, comprising: Acquire point cloud data of the area to be detected; An initial detection box is identified from the point cloud data of the region to be detected based on the detection network; The initial detection box is adjusted to a target detection box according to the adjustment parameters. The adjustment parameters are determined based on the parameter values ​​of all detection boxes in at least two different detection regions. The detection boxes of the same type of object in the different detection regions are obtained by identifying a sample dataset. The sample dataset contains at least point cloud sample data in the different detection regions. The two different detection regions include a region to be detected and a detection region. The region to be detected is a target domain training set, and the detection region is a source domain training set. There is a domain gap between the sample data in the target domain training set and the sample data in the source domain training set. The domain gap is used to characterize the differences between the sample data in the target domain training set and the sample data in the source domain training set.

9. A processing device for a detection box of an object, comprising: The first acquisition unit is used to acquire a sample dataset, wherein the sample dataset contains at least point cloud sample data from two different detection regions, wherein the two different detection regions include a region to be detected and a detection region, the region to be detected is a target domain training set, the detection region is a source domain training set, and there is a domain gap between the sample data in the target domain training set and the sample data in the source domain training set, the domain gap being used to characterize the differences between the sample data in the target domain training set and the sample data in the source domain training set; The first identification unit is used to identify the sample dataset and obtain detection boxes of the same type of objects located in the different detection areas; The second acquisition unit is used to acquire the parameter values ​​of all detection boxes in the different detection areas respectively, wherein the objects corresponding to all detection boxes are of the same type. The determining unit is used to determine the adjustment parameters based on the parameter values ​​of all detection frames in the different detection areas; The first adjustment unit is used to adjust the initial detection boxes of the same type of object output by the detection network according to the adjustment parameters to obtain the target detection box, wherein the initial detection box is identified by the detection network from the point cloud data of the region to be detected in the different detection regions; The point cloud sample data within the detection area is used to train the detection network. The second acquisition unit is further used to acquire the first parameter values ​​of all detection boxes within the detection area; acquire the second parameter values ​​of all detection boxes within the area to be detected; and determine the adjustment parameters based on the first parameter values ​​and the second parameter values. The second acquisition unit is further configured to determine the adjustment parameter based on the first parameter value and the second parameter value by: comparing the second parameter value and the first parameter value to obtain a comparison result; and determining the adjustment parameter based on the comparison result.

10. The apparatus according to claim 9, wherein, The determining unit includes: The first determining submodule is used to determine the adjustment parameter of the target parameter type based on the first parameter value of the target parameter type and the second parameter value of the target parameter type, wherein the adjustment parameter is used to adjust the target size of the initial detection box, and the type of the target size is the target parameter type.

11. The apparatus according to claim 9, wherein, The first adjustment unit includes: An adjustment module is used to adjust the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box, wherein the target detection box includes the adjusted size of the initial detection box and / or the adjusted point cloud data within the initial detection box.

12. The apparatus according to claim 11, wherein, The first adjustment unit includes: The scaling operation submodule is used to scale the size of the initial detection box and / or the point cloud data within the initial detection box according to the adjustment parameters to obtain the target detection box.

13. A processing apparatus for an object detection frame, comprising: The third acquisition unit is used to acquire point cloud data of the area to be detected. The second recognition unit is used to identify an initial detection box from the point cloud data of the area to be detected based on the detection network; The second adjustment unit is used to adjust the initial detection box into a target detection box according to adjustment parameters. The adjustment parameters are determined based on the parameter values ​​of all detection boxes in at least two different detection regions. The detection boxes of the same type of object in the different detection regions are obtained by identifying a sample dataset. The sample dataset contains at least point cloud sample data in the different detection regions. The two different detection regions include a region to be detected and a detection region. The region to be detected is a target domain training set, and the detection region is a source domain training set. There is a domain gap between the sample data in the target domain training set and the sample data in the source domain training set. The domain gap is used to characterize the difference between the sample data in the target domain training set and the sample data in the source domain training set.

14. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

15. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

16. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images