Object Detection Method, Computer-Readable Storage Medium, and Driving Device

Abstract

The present invention relates to the field of autonomous driving technology, and specifically provides a target detection method, a computer-readable storage medium, and a driving device, aiming to solve the problem that when performing target detection based on point cloud data, some objects with reflective characteristics are easily misdetected as targets. For this purpose, the target detection method of the present invention includes: obtaining point cloud data of the area to be detected, performing target detection based on the point cloud data to obtain an initial detection result, the initial detection result including at least one initial detection box corresponding to the target to be detected, determining the tracking speed of the initial detection box, and correcting the initial detection result according to the tracking speed in combination with the reflection intensity of the point cloud in the initial detection box to obtain the final target detection result. This method can effectively avoid the problem that some objects with reflective characteristics are easily misdetected as targets when performing target detection based on point cloud data in the prior art, and effectively improve the accuracy of target detection.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and specifically provides a target detection method, a computer-readable storage medium, and a driving device. Background Art

[0002] In the prior art, target detection is usually performed by acquiring point cloud data of an interested region containing a target, determining point clouds whose geometric shapes are similar to the target shape based on the point cloud data, and outputting a target detection box based on the contour of the point cloud. Taking vehicle recognition in the driving process as an example, by collecting point cloud data of an interested region, point clouds whose geometric shapes are similar to a vehicle are determined based on the point cloud data and a vehicle detection box is output. However, when using the existing method for target detection, there will be a problem that some traffic signs with reflective materials are misdetected as vehicles, resulting in low target detection accuracy. Summary of the Invention

[0003] The present invention aims to solve the above technical problems, that is, to solve the problem that when performing target detection based on point cloud data in the prior art, some objects with reflective characteristics are easily misdetected as targets.

[0004] In a first aspect, the present invention provides a target detection method, which includes:

[0005] Acquire point cloud data of a region to be detected, where the point cloud data includes the reflection intensity of the point cloud;

[0006] Perform target detection based on the point cloud data to obtain an initial detection result, where the initial detection result includes at least one initial detection box corresponding to a target to be detected;

[0007] Determine the tracking speed of the initial detection box;

[0008] According to the tracking speed and the reflection intensity of the point cloud within the initial detection box, correct the initial detection result to obtain a target detection result.

[0009] In some embodiments, the step of correcting the initial detection result according to the tracking speed and the reflection intensity of the point cloud within the initial detection box to obtain a target detection result includes:

[0010] According to the number of points with a reflection intensity greater than a reflection intensity threshold within the initial detection box and the total number of points in the point cloud within the initial detection box, determine the high-reflection-intensity point ratio of the initial detection box;

[0011] According to the tracking speed, a speed threshold, the high-reflection-intensity point ratio, and a high-reflection-intensity point ratio threshold, correct the initial detection result to obtain the target detection result.

[0012] In some embodiments, correcting the initial detection result according to the tracking speed, speed threshold, high-reflection intensity point ratio, and high-reflection intensity point ratio threshold to obtain a target detection result includes:

[0013] According to the tracking speed, the speed threshold, the high-reflection intensity point ratio, and the high-reflection intensity point ratio threshold, removing the initial detection boxes corresponding to the tracking speed being less than the speed threshold and the high-reflection intensity point ratio being greater than the high-reflection intensity point ratio threshold.

[0014] In some embodiments, performing target detection based on the point cloud data to obtain an initial detection result, where the initial detection result includes at least one initial detection box corresponding to a target to be detected, includes:

[0015] Performing target detection on the point cloud data using a three-dimensional target detection network model to obtain at least one initial detection box corresponding to the target to be detected, and the initial detection box is a three-dimensional initial detection box.

[0016] In some embodiments, determining the tracking speed of the initial detection box includes:

[0017] Tracking the initial detection box, and obtaining the displacement of the initial detection box in two frames of point cloud data at a preset frame interval and the time interval between the two frames of point cloud data;

[0018] Determining the tracking speed of the initial detection box according to the displacement and the time interval.

[0019] In some embodiments, tracking the initial detection box and obtaining the displacement of the initial detection box in two frames of point cloud data at a preset frame interval includes:

[0020] Obtaining the position information of the initial detection box in two frames of point cloud data at a preset frame interval and converting the position information of the initial detection box in the two frames of point cloud data to the same coordinate system to obtain the converted position information of the initial detection box in the two frames of point cloud data;

[0021] According to the converted position information of the initial detection box, calculating the intersection over union of any two initial detection boxes in the two frames of point cloud data, and determining two associated initial detection boxes in the two frames of point cloud data according to the calculation result;

[0022] Determining the displacement of the initial detection box according to the converted position information of the two associated initial detection boxes.

[0023] In some embodiments, obtaining the point cloud data of the area to be detected includes:

[0024] Obtain the point cloud data of the area to be detected based on lidar.

[0025] In a second aspect, the present invention provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the object detection method described in any one of the above is implemented.

[0026] In a third aspect, the present invention provides an electronic device, which includes a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, the object detection method described in any one of the above is implemented.

[0027] In a fourth aspect, the present invention provides a driving device, which includes a driving device body, a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, the object detection method described in any one of the above is implemented.

[0028] In the case of adopting the above technical solutions, the present invention can obtain the point cloud data of the area to be detected, perform object detection based on the point cloud data of the area to be detected to obtain an initial detection result, and the initial detection result includes at least one initial detection box corresponding to the object to be detected. By determining the tracking speed of the initial detection box and correcting the initial detection result according to the tracking speed in combination with the reflection intensity of the point cloud in the initial detection box, the final object detection result can be obtained, thereby effectively avoiding the problem in the prior art that when performing object detection based on point cloud data, some objects with reflective characteristics are easily misdetected as objects, and effectively improving the accuracy of object detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The following describes the preferred embodiments of the present invention with reference to the drawings, in which:

[0030] Figure 1 is a flowchart of an object detection method provided by an embodiment of the present invention;

[0031] Figure 2 is a flowchart of a tracking speed determination method provided by an embodiment of the present invention;

[0032] Figure 3 is a flowchart of a method for correcting the initial detection result provided by an embodiment of the present invention;

[0033] Figure 4 is a schematic diagram of a driving environment provided by the present invention;

[0034] Figure 5 is a schematic diagram of a vehicle detection result based on the initial detection result provided by the present invention;

[0035] Figure 6 It is a schematic diagram of the vehicle detection result of the corrected driving environment provided by the present invention. Specific Embodiments

[0036] Some embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0037] In the prior art, target detection is usually carried out by acquiring the point cloud data of the region of interest containing the target, determining the point cloud whose geometric shape is similar to the target shape according to the point cloud data, and outputting the target detection frame based on the contour of the point cloud. Taking vehicle recognition in the driving process as an example, by collecting the point cloud data of the region of interest, the point cloud whose geometric shape is similar to the vehicle is determined according to the point cloud data and the vehicle detection frame is output. However, when using the existing method for target detection, for some traffic signs with reflective materials, the corresponding point cloud contour will spread around, forming a point cloud geometric shape larger than the real object, resulting in the point cloud geometric shape being similar to that of a vehicle, so it is easy to misdetect the traffic sign as a vehicle.

[0038] In view of this, the present invention provides a target detection method, which includes acquiring the point cloud data of the region to be detected, performing target detection based on the point cloud data of the region to be detected to obtain an initial detection result, where the initial detection result includes at least one initial detection frame corresponding to the target to be detected, determining the tracking speed of the initial detection frame, and correcting the initial detection result according to the tracking speed in combination with the reflection intensity of the point cloud in the initial detection frame to obtain the final target detection result. This method can effectively avoid the problem in the prior art that when performing target detection based on point cloud data, some objects with reflective characteristics are easily misdetected as targets according to the point cloud geometric shape, and effectively improves the accuracy of target detection.

[0039] See Figure 1 as shown Figure 1 It is a schematic flowchart of a target detection method provided by an embodiment of the present invention, which may include:

[0040] Step S11: Acquire the point cloud data of the region to be detected, where the point cloud data includes the reflection intensity of the point cloud;

[0041] Step S12: Perform target detection based on the point cloud data to obtain an initial detection result, where the initial detection result includes at least one initial detection frame corresponding to the target to be detected;

[0042] Step S13: Determine the tracking speed of the initial detection frame;

[0043] Step S14: Modify the initial detection result according to the tracking speed and the reflection intensity of the point cloud within the initial detection box to obtain the target detection result.

[0044] In some embodiments, step S11 may specifically be to obtain point cloud data of the area to be detected based on a lidar. Using a lidar is conducive to obtaining high-precision point cloud data.

[0045] In some other embodiments, point cloud data of the area to be detected may also be obtained based on a millimeter-wave radar.

[0046] In some embodiments, the point cloud data may at least include the reflection intensity of the point cloud, where each point in the point cloud corresponds to a reflection intensity.

[0047] In some other embodiments, the point cloud data may further include the three-dimensional coordinate data and / or color data of each point.

[0048] In some embodiments, step S12 may specifically be to perform target detection on the point cloud data using a three-dimensional target detection network model to obtain at least one initial detection box corresponding to the target to be detected. The initial detection box is a three-dimensional initial detection box. Performing target detection using a three-dimensional target detection network can avoid losing the information of the original data and is conducive to improving the detection accuracy.

[0049] As an example, the three-dimensional target detection network model may include a PointPillar model, a VoxelNet model, or a CenterPoints model.

[0050] In some other embodiments, the initial detection result may also be obtained by first clustering the point cloud data and then using a classification model for classification.

[0051] In some embodiments, referring to Figure 2 shown, Figure 2 is a schematic flowchart of the tracking speed determination method provided by an embodiment of the present invention. Step S13 may specifically be:

[0052] Step S131: Track the initial detection box, and obtain the displacement of the initial detection box in two frames of point cloud data at a preset frame interval and the time interval between the two frames of point cloud data;

[0053] Step S132: Determine the tracking speed of the initial detection box according to the displacement and the time interval.

[0054] In some embodiments, when tracking the initial detection box in step S131 to obtain the displacement of the initial detection box in two frames of point cloud data at a preset frame interval, it may specifically be:

[0055] Obtain the position information of the initial detection boxes in two frames of point cloud data separated by a preset number of frames, and convert the position information of the initial detection boxes in the two frames of point cloud data to the same coordinate system to obtain the converted position information of the initial detection boxes in the two frames of point cloud data;

[0056] According to the converted position information of the initial detection boxes, calculate the intersection over union (IoU) of any two initial detection boxes in the two frames of point cloud data, and determine two associated initial detection boxes in two adjacent frames of point cloud data based on the calculation result;

[0057] Determine the displacement of the initial detection box according to the converted position information of the two associated initial detection boxes.

[0058] Among them, the preset number of frames can be set as needed. As an example, it is possible to obtain that the preset number of frames is zero, that is, two adjacent frames of point cloud data are used to determine the tracking speed. The following will be described based on two adjacent frames of point cloud data as an example.

[0059] In some embodiments, by performing object detection on two adjacent frames of point cloud data, the position information of the initial detection boxes corresponding to the target to be detected in each frame of point cloud data can be obtained respectively.

[0060] Among them, the position information of the initial detection box is the position information of the initial detection box in the vehicle coordinate system. The vehicle coordinate systems corresponding to the initial detection boxes in different frames of point cloud data are different. Therefore, before calculating the intersection over union (IoU) of any two initial detection boxes in the two frames of point cloud data, it is necessary to convert the position information of the initial detection boxes in the two frames of point cloud data to the same coordinate system.

[0061] In some embodiments, converting the position information of the initial detection boxes in the two frames of point cloud data to the same coordinate system can be as follows:

[0062] Construct a three-dimensional coordinate system with the position of the vehicle at startup as the origin. The three-dimensional coordinate system takes the forward direction of the vehicle as the positive x-axis direction, the direction perpendicular to the x-axis direction and pointing to the left side of the vehicle as the y-axis direction, and the direction perpendicular to the plane where the x-axis and y-axis are located and pointing to the roof of the vehicle as the z-axis direction;

[0063] Obtain the first position information of the vehicle in the three-dimensional coordinate system when collecting the first frame of point cloud data and the second position information of the vehicle in the three-dimensional coordinate system when obtaining the second frame of point cloud data adjacent to the first frame;

[0064] According to the first position information and the position information of the initial detection box in the first frame of point cloud data in the vehicle coordinate system corresponding to when collecting the first frame of point cloud data, calculate the position information of the initial detection box in the first frame of point cloud data after conversion to the three-dimensional coordinate system;

[0065] Calculate the position information of the initial detection box in the second frame of point cloud data after being transformed into the three-dimensional coordinate system according to the second position information and the position information of the initial detection box in the vehicle-mounted coordinate system corresponding to the acquisition of the second frame of point cloud data; thereby, transform the position information of the initial detection box in the two frames of point cloud data into the same coordinate system, and the intersection over union of the two initial detection boxes and the displacement of the initial detection box can be calculated based on the position information of the initial detection box in the first frame of point cloud data after being transformed into the three-dimensional coordinate system and the position information of the initial detection box in the second frame of point cloud data after being transformed into the three-dimensional coordinate system.

[0066] By calculating the intersection over union of any two initial detection boxes in two adjacent frames of point cloud data, two associated initial detection boxes in two adjacent frames of point cloud data can be determined when the calculation result meets the preset condition, and the two associated initial detection boxes correspond to the same target to be detected at different times. In some embodiments, the preset condition may be that the intersection over union of the two initial detection boxes is greater than the intersection over union threshold.

[0067] In some other embodiments, the initial detection box can also be tracked based on the Kalman filtering algorithm to obtain the position information of the initial detection box in two adjacent frames of point cloud data.

[0068] In some embodiments, as shown in Figure 3 shown, Figure 3 is a schematic flowchart of the method for correcting the initial detection result provided by the embodiment of the present invention. Step S14 may specifically be:

[0069] Step S141: Determine the high-reflection-intensity point ratio of the initial detection box according to the number of points with a reflection intensity greater than the reflection intensity threshold in the initial detection box and the total number of points in the point cloud within the initial detection box;

[0070] Step S142: Correct the initial detection result according to the tracking speed, speed threshold, high-reflection-intensity point ratio, and high-reflection-intensity point ratio threshold to obtain the target detection result.

[0071] Among them, the speed threshold and the high-reflection-intensity point ratio threshold can be flexibly set according to actual application requirements.

[0072] In some embodiments, step S142 may specifically be: Remove the initial detection box corresponding to a tracking speed less than the speed threshold and a high-reflection-intensity point ratio greater than the high-reflection-intensity point ratio threshold according to the tracking speed, speed threshold, high-reflection-intensity point ratio, and high-reflection-intensity point ratio threshold. This method can distinguish real targets and pseudo targets from two aspects of motion characteristics and reflection intensity distribution characteristics by combining the tracking speed with the reflection intensity of the point cloud within the initial detection box, and then correct the initial detection result to obtain the final target detection result.

[0073] The above is a target detection method provided by an embodiment of the present invention, which includes obtaining point cloud data of a region to be detected, performing target detection based on the point cloud data to obtain an initial detection result, where the initial detection result includes at least one initial detection box corresponding to a target to be detected, determining the tracking speed of the initial detection box, and correcting the initial detection result according to the tracking speed in combination with the reflection intensity of the point cloud within the initial detection box to obtain a final target detection result. This method can effectively avoid the problem in the prior art that when performing target detection based on point cloud data, some objects with reflective characteristics are easily misdetected as targets, and effectively improves the accuracy of target detection.

[0074] As an example, the target detection method provided by an embodiment of the present invention can be applied to vehicle detection in autonomous driving.

[0075] See Figure 4 shown in Figure 4 is a schematic diagram of a driving environment provided by the present invention. The driving environment includes vehicles and traffic markers in the middle of the road with reflective materials attached. Applying the target detection method provided by an embodiment of the present invention to Figure 4 the vehicle detection in the shown driving environment may include:

[0076] Obtain the point cloud data of the driving environment, where the point cloud data includes the reflection intensity of the point cloud;

[0077] Use a vehicle detection model to detect the obtained point cloud data and obtain an initial detection result;

[0078] If directly outputting based on the initial detection result, as Figure 5 shown, since the point cloud contour of the obtained traffic marker has expanded relative to the true contour of the traffic marker, and the expanded point cloud shape is similar to that of a vehicle, the traffic marker is also misdetected as a vehicle in the initial detection result and the corresponding initial detection box is output;

[0079] Determine the tracking speed of the initial detection box;

[0080] Count the number of points with a reflection intensity greater than the reflection intensity threshold within the initial detection box, and determine the high-reflection-intensity point ratio within the initial detection box according to the number of points and the total number of points in the point cloud within the initial detection box;

[0081] According to the tracking speed, the set speed threshold, the high-reflection-intensity point ratio, and the set high-reflection-intensity point ratio threshold, remove the initial detection box corresponding to the tracking speed less than the speed threshold and the high-reflection-intensity point ratio greater than the high-reflection-intensity point ratio threshold, and then filter out the initial detection box corresponding to the traffic marker, and further obtain and output the corrected target detection result, as Figure 6 shown in Figure 6It is a schematic diagram of the vehicle detection result of the corrected driving environment provided by the present invention.

[0082] By applying the object detection method provided by the embodiments of the present invention to Figure 4 perform vehicle detection on the driving environment shown in, it is possible to effectively distinguish traffic signs that are static and pasted with reflective materials from vehicles by combining the tracking speed with the reflection intensity of the point cloud within the initial detection frame and correct the initial detection result, overcoming the problem in the prior art that some traffic signs with reflective characteristics are easily misdetected as vehicles according to the geometric shape of the point cloud, and effectively improving the accuracy of object detection.

[0083] On the other hand, the present invention also provides a computer-readable storage medium in which a computer program is stored. When the computer program is executed by a processor, it can implement the object detection method in any of the above embodiments. The computer-readable storage medium may be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiments of the present invention is a non-transitory computer-readable storage medium.

[0084] On the other hand, the present invention also provides an electronic device, which includes: a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, it implements the object detection method described in any of the above embodiments.

[0085] On the other hand, the present invention also provides a driving device, which includes a driving device body, a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, it implements the object detection method described in any of the above embodiments.

[0086] In some embodiments, the driving device may further include a lidar for acquiring point cloud data of the area to be detected.

[0087] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A target detection method, characterized in that, it includes: Obtain the point cloud data of the area to be detected, where the point cloud data includes the reflection intensity of the point cloud; Perform target detection based on the point cloud data to obtain an initial detection result, where the initial detection result includes at least one initial detection box corresponding to the target to be detected; Determine the tracking speed of the initial detection box; According to the tracking speed and the reflection intensity of the point cloud within the initial detection box, correct the initial detection result to obtain a target detection result, where, according to the tracking speed and the reflection intensity of the point cloud within the initial detection box, correcting the initial detection result to obtain a target detection result includes: Determine the high-reflection-intensity point ratio of the initial detection box according to the number of points with a reflection intensity greater than the reflection intensity threshold within the initial detection box and the total number of points in the point cloud within the initial detection box; According to the tracking speed, the speed threshold, the high-reflection-intensity point ratio, and the high-reflection-intensity point ratio threshold, remove the initial detection box corresponding to the case where the tracking speed is less than the speed threshold and the high-reflection-intensity point ratio is greater than the high-reflection-intensity point ratio threshold.

2. The method according to claim 1, characterized in that, The performing target detection based on the point cloud data to obtain an initial detection result, where the initial detection result includes at least one initial detection box corresponding to the target to be detected, includes: Perform target detection on the point cloud data using a three-dimensional target detection network model to obtain at least one initial detection box corresponding to the target to be detected, and the initial detection box is a three-dimensional initial detection box.

3. The method according to claim 1, characterized in that, The determining the tracking speed of the initial detection box includes: Track the initial detection box, and obtain the displacement of the initial detection box in two frames of point cloud data with a preset frame interval and the time interval between the two frames of point cloud data; Determine the tracking speed of the initial detection box according to the displacement and the time interval.

4. The method according to claim 3, characterized in that, The tracking the initial detection box and obtaining the displacement of the initial detection box in two frames of point cloud data with a preset frame interval includes: Obtain the position information of the initial detection box in two frames of point cloud data with a preset frame interval and convert the position information of the initial detection box in the two frames of point cloud data to the same coordinate system to obtain the converted position information of the initial detection box in the two frames of point cloud data; According to the converted position information of the initial detection box, calculate the intersection over union of any two initial detection boxes in the two frames of point cloud data, and determine the two associated initial detection boxes in the two frames of point cloud data according to the calculation result; Determine the displacement of the initial detection box according to the converted position information of the two associated initial detection boxes.

5. The method according to claim 1, characterized in that, The obtaining the point cloud data of the area to be detected includes: Obtain the point cloud data of the area to be detected based on a lidar.

6. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the object detection method according to any one of claims 1 to 5.

7. An electronic device, characterized in that it includes a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, the object detection method according to any one of claims 1 to 5 is implemented.

8. A driving device, characterized in that it includes a driving device body, a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, the object detection method according to any one of claims 1 to 5 is implemented.

Images