Object detection task BEV feature extraction method and system based on multi-scale ROI projection

The invention discloses an object detection task BEV feature extraction method and system based on multi-scale ROI projection, and belongs to the technical field of machine vision. The method comprises the steps of performing conversion operation on an original point cloud to obtain an enhanced point cloud, and performing BEV feature conversion on the original point cloud and the enhanced point cloud by using a 3D backbone network; dividing the original BEV feature map and the enhanced BEV feature map into a plurality of sub-regions according to the target scale of the to-be-detected object; calculating the ROI score of each sub-region, determining ROI candidate sub-regions based on the ROI scores, and determining the corresponding BEV features as contrast learning examples; and pre-training an object detection model by using the comparison instance, and optimizing and updating the pre-trained model. According to the method, the ROI in the BEV space can be more related to the object detection task, the traini