3D target detection algorithm based on combination of time attention and deformable cross attention

The invention relates to the technical field of 3D target detection, in particular to a 3D target detection algorithm based on combination of time attention and deformable cross attention, and the algorithm comprises the following steps: S1, data construction; step S2, data preprocessing; s3, a time attention module; according to the method, the learnable parameter BEVQuery with the dimension of 200 * 200 is initialized, information of a historical frame and information of a current frame are fused, feature extraction of six input images is carried out, a network architecture of Resnet101 is adopted, feature fusion is carried out, then the deformable cross attention is expanded to 3D perception, and the method has the advantages that the learning parameter BEVQuery with the dimension of 200 * 200 is optimized; the method comprises the following steps of: performing dimension raising on each BEV Query on a BEV plane, setting four heights on the dimension of height, mapping each reference point to an image coor