Deep Human Detection Based On Double Human Deformable Part Models

Human detection has abroad application foreground in driver assistance system and intelligent monitoring system, and it has gradually become an important research content in the fields of computer vision and artificial intelligence. At present, human detection technology still has much space for improvement. The purpose of this paper is to research and improve the two problems of the current algorithm, the first is easily miss-detection when detecting multi human in the very near distance or with occlusion, and the second is how to extract the better human features.Firstly, based on the detailed analysis of DPM, this paper proposes a new algorithm based on double human DPM. The matching fusion method for regions is used to reduce missed detection of multi human detection. Then this paper deeply studies the deep model and proposes human detection based on double human deep model. Double human deep model is used to improve double human DPM by confirming detection windows from double human DPM and sliding windows from the input image. This can make full use of their advantages and make up for their shortcomings.From the experiment, it can be seen that double human DPM reduce miss rate of detection window to a certain extent, and double human deep model reduce false rate of detection window further. The combination of both improves detection efficiency and detection effect. The results show that the algorithm achieved the expected detection.