Monocular pedestrian 3D positioning method based on information fusion

The invention provides a monocular pedestrian 3D positioning method based on information fusion. According to the method, a complex pedestrian positioning problem is decomposed to obtain three parts,namely fuzzy positioning, depth information reconstruction and positioning optimization based on information fusion. Specifically, firstly, in the fuzzy positioning stage, human skeleton key points are obtained from an original image, and pedestrian fuzzy positions are obtained according to the triangle similarity theorem; secondly, in a depth information reconstruction stage, depth estimation andsemantic segmentation is conducted on the original image in a parallel network to obtain an aerial view with depth information; and finally, accurate positioning is carried out through a positioningoptimization method based on information fusion. According to the method, the accurate position of the pedestrian can be obtained; an experimental result on a KITTI data set also shows superiority ofthe method. 本发明提出了一种基于信息融合的单