Vehicle information detection method, method for training detection model, electronic device and storage medium
A vehicle information detection method, a method for training a detection model, an electronic device and a storage medium are provided, and relates to the technical field of artificial intelligence, in particular to the technical field of computer vision and deep learning. The method includes: performing a first target detection operation based on an image of a target vehicle, to obtain a first detection result for target information of the target vehicle; performing an error detection operation based on the first detection result, to obtain error information; and performing a second target detection operation based on the first detection result and the error information, to obtain a second detection result for the target information.
1. A vehicle information detection method, comprising:
performing a first target detection operation based on an image of a target vehicle, to obtain a first detection result for target information of the target vehicle;
performing an error detection operation based on the first detection result, to obtain error information; and
performing a second target detection operation based on the first detection result and the error information, to obtain a second detection result for the target information;
fusing the first detection result and the second detection result based on the error information, to obtain a third detection result for the target information;
wherein the fusing the first detection result and the second detection result based on the error information, to obtain the third detection result for the target information comprises:
obtaining a first weight negatively correlated with the error information and a second weight positively correlated with the error information based on the error information;
taking the first weight as a weight of the first detection result;
taking the second weight as a weight of the second detection result; and
performing weighted addition on the first detection result and the second detection result based on the weight of the first detection result and the weight of the second detection result, to obtain the third detection result for the target information.
2. The method of claim 1 , wherein the first detection result comprises a target feature map corresponding to the image, and the error information comprises an error prediction map; and
the performing the second target detection operation based on the first detection result and the error information, to obtain the second detection result for the target information comprises:
performing channel connection on the target feature map corresponding to the image and the error prediction map to obtain an input feature map; and
performing the second target detection operation on the input feature map to obtain the second detection result for the target information.
3. The method of claim 1 , wherein the target information comprises an offset of a bottom surface center point of the target vehicle; and
the method further comprises:
obtaining a ground equation and a two-dimensional detection box of the target vehicle in the image;
obtaining coordinates of a projection point of the bottom surface center point in the image according to the third detection result of the offset and the two-dimensional detection box;
obtaining depth information of the bottom surface center point according to the coordinates of the projection point and the ground equation; and
obtaining a detection result for an actual position of the bottom surface center point according to the coordinates of the projection point, the depth information, and a camera parameter corresponding to the image.
4. The method of claim 3 , further comprising:
obtaining height information of the target vehicle; and
obtaining a detection result for an actual position of the target vehicle according to the detection result for the actual position of the bottom surface center point and the height information of the target vehicle.
5. A method for training a vehicle information detection model, comprising:
obtaining target information of a plurality of vehicles based on radar point cloud data of the plurality of vehicles; and
training based on the target information of the plurality of vehicles and images of the plurality of vehicles to obtain a vehicle detection model, wherein the vehicle detection model is used for obtaining a detection result for target information of a target vehicle based on an image of the target vehicle;
wherein the obtaining target information of the plurality of vehicles based on the radar point cloud data of the plurality of vehicles comprises:
obtaining three-dimensional detection boxes of the plurality of vehicles in a radar coordinate system based on the radar point cloud data of the plurality of vehicles;
calculating coordinates of projection points of respective vertexes and bottom surface center points of the three-dimensional detection boxes of the plurality of vehicles in the images according to the three-dimensional detection boxes of the plurality of vehicles in the radar coordinate system, a pre-configured radar parameter and a camera parameter;
obtaining two-dimensional detection boxes of the plurality of vehicles according to the coordinates of the projection points of the respective vertexes of the three-dimensional detection boxes of the plurality of vehicles in the images; and
obtaining offsets of the bottom surface center points of the plurality of vehicles according to the coordinates of the projection points of the bottom surface center points of the three-dimensional detection boxes of the plurality of vehicles in the images and the two-dimensional detection boxes of the plurality of vehicles.
6. An electronic device, 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, and the instruction is executed by the at least one processor to enable the at least one processor to perform operations of:
performing a first target detection operation based on an image of a target vehicle, to obtain a first detection result for target information of the target vehicle;
performing an error detection operation based on the first detection result, to obtain error information; and
performing a second target detection operation based on the first detection result and the error information, to obtain a second detection result for the target information;
fusing the first detection result and the second detection result based on the error information, to obtain a third detection result for the target information;
wherein the fusing the first detection result and the second detection result based on the error information, to obtain the third detection result for the target information comprises:
obtaining a first weight negatively correlated with the error information and a second weight positively correlated with the error information based on the error information;
taking the first weight as a weight of the first detection result;
taking the second weight as a weight of the second detection result; and
performing weighted addition on the first detection result and the second detection result based on the weight of the first detection result and the weight of the second detection result, to obtain the third detection result for the target information.
7. The electronic device of claim 6 , wherein the first detection result comprises a target feature map corresponding to the image, and the error information comprises an error prediction map; and
the performing the second target detection operation based on the first detection result and the error information, to obtain the second detection result for the target information comprises:
performing channel connection on the target feature map corresponding to the image and the error prediction map to obtain an input feature map; and
performing the second target detection operation on the input feature map to obtain the second detection result for the target information.
8. The electronic device of claim 6 , wherein the target information comprises an offset of a bottom surface center point of the target vehicle; and
wherein the instruction is executed by the at least one processor to enable the at least one processor to further perform operations of:
obtaining a ground equation and a two-dimensional detection box of the target vehicle in the image;
obtaining coordinates of a projection point of the bottom surface center point in the image according to the third detection result of the offset and the two-dimensional detection box;
obtaining depth information of the bottom surface center point according to the coordinates of the projection point and the ground equation; and
obtaining a detection result for an actual position of the bottom surface center point according to the coordinates of the projection point, the depth information, and a camera parameter corresponding to the image.
9. The electronic device of claim 8 , wherein the instruction is executed by the at least one processor to enable the at least one processor to further perform operations of:
obtaining height information of the target vehicle; and
obtaining a detection result for an actual position of the target vehicle according to the detection result for the actual position of the bottom surface center point and the height information of the target vehicle.
10. An electronic device, 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, and the instruction is executed by the at least one processor to enable the at least one processor to perform operations of:
obtaining target information of a plurality of vehicles based on radar point cloud data of the plurality of vehicles; and
training based on the target information of the plurality of vehicles and images of the plurality of vehicles to obtain a vehicle detection model, wherein the vehicle detection model is used for obtaining a detection result for target information of a target vehicle based on an image of the target vehicle;
wherein the obtaining target information of the plurality of vehicles based on the radar point cloud data of the plurality of vehicles comprises:
obtaining three-dimensional detection boxes of the plurality of vehicles in a radar coordinate system based on the radar point cloud data of the plurality of vehicles;
calculating coordinates of projection points of respective vertexes and bottom surface center points of the three-dimensional detection boxes of the plurality of vehicles in the images according to the three-dimensional detection boxes of the plurality of vehicles in the radar coordinate system, a pre-configured radar parameter and a camera parameter;
obtaining two-dimensional detection boxes of the plurality of vehicles according to the coordinates of the projection points of the respective vertexes of the three-dimensional detection boxes of the plurality of vehicles in the images; and
obtaining offsets of the bottom surface center points of the plurality of vehicles according to the coordinates of the projection points of the bottom surface center points of the three-dimensional detection boxes of the plurality of vehicles in the images and the two-dimensional detection boxes of the plurality of vehicles.
11. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions cause a computer to perform operations of:
performing a first target detection operation based on an image of a target vehicle, to obtain a first detection result for target information of the target vehicle;
performing an error detection operation based on the first detection result, to obtain error information; and
performing a second target detection operation based on the first detection result and the error information, to obtain a second detection result for the target information;
fusing the first detection result and the second detection result based on the error information, to obtain a third detection result for the target information;
wherein the fusing the first detection result and the second detection result based on the error information, to obtain the third detection result for the target information comprises:
obtaining a first weight negatively correlated with the error information and a second weight positively correlated with the error information based on the error information;
taking the first weight as a weight of the first detection result;
taking the second weight as a weight of the second detection result; and
performing weighted addition on the first detection result and the second detection result based on the weight of the first detection result and the weight of the second detection result, to obtain the third detection result for the target information.
12. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions cause a computer to perform operations of:
obtaining target information of a plurality of vehicles based on radar point cloud data of the plurality of vehicles; and
training based on the target information of the plurality of vehicles and images of the plurality of vehicles to obtain a vehicle detection model, wherein the vehicle detection model is used for obtaining a detection result for target information of a target vehicle based on an image of the target vehicle;
wherein the obtaining target information of the plurality of vehicles based on the radar point cloud data of the plurality of vehicles comprises:
obtaining three-dimensional detection boxes of the plurality of vehicles in a radar coordinate system based on the radar point cloud data of the plurality of vehicles;
calculating coordinates of projection points of respective vertexes and bottom surface center points of the three-dimensional detection boxes of the plurality of vehicles in the images according to the three-dimensional detection boxes of the plurality of vehicles in the radar coordinate system, a pre-configured radar parameter and a camera parameter;
obtaining two-dimensional detection boxes of the plurality of vehicles according to the coordinates of the projection points of the respective vertexes of the three-dimensional detection boxes of the plurality of vehicles in the images; and
obtaining offsets of the bottom surface center points of the plurality of vehicles according to the coordinates of the projection points of the bottom surface center points of the three-dimensional detection boxes of the plurality of vehicles in the images and the two-dimensional detection boxes of the plurality of vehicles.