Visual compiler-based person and pose detection method

Abstract

The invention provides a visual compiler-based person and pose detection method. The method mainly comprises the steps of performing data synthesis of scene description; learning a net through synthesized data; defining the net by adopting basic blocks; and performing joint locating by using a pose net. According to the process, the scene description is used as an input of a visual compiler, a pedestrian detection system and a pose estimation system are trained by calibrated real data annotations, and the net is learnt through the synthesized data; then the net is defined by using two basic units including a residual module and a spatial belief module; and finally pedestrians are located by using the pose net. According to the method, the detected annotations, body part positions and segmented mask codes can be automatically obtained, the pedestrians are located by using a video camera, and the pedestrians are subjected to pose estimation and activity analysis; and the influence of illumination, shielding and the like on the detection is reduced, so that the identification efficiency is effectively improved.

Description

technical field [0001] The invention relates to the field of person posture detection, in particular to a method for person and posture detection based on a visual compiler. Background technique [0002] The detection of human motion and posture has become a research hotspot in the field of computer vision due to its wide application in video surveillance, virtual reality, and interpersonal intelligent interaction. It can be used for intelligent surveillance in public places and monitoring of dangerous postures in crowds. Although the research on human body posture detection in China has made important progress in recent years, the high complexity and variability of human body posture make the accuracy and efficiency of recognition not fully meet the requirements of related industries. Different lighting, viewing angles, and background conditions will cause differences in human behavior in posture and characteristics. In addition, human body self-occlusion, partial occlusion...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that lighting, occlusion, etc. will have an impact, the purpose of the present invention is to provide a visual compiler-based character and posture detection method, first use the scene description as the input of the visual compiler, and use the calibrated real data annotations to train Pedestrian detection system and pose estimation system, and then learn the network through synthetic data; then use the two basic units of residual module and spatial confidence module to define the network, and finally use the pose network to locate pedestrians

Images