论文部分内容阅读
Traditional human detection using pre-trained detectors tends to be computationally intensive for time-critical tracking tasks, and the detection rate is prone to be unsatisfying when occlusion, motion blur and body deformation occur frequently. A spat