Invigilator tracking method based on postures

Abstract

The invention discloses an invigilator tracking method based on postures, which comprises the following steps: determining the positions of different persons by analyzing the relation between joint points of the upper body part of a human body, obtaining the positions of invigilators by using human body joint point data generated by deep learning, extracting the characteristics of the invigilators, and matching front and rear frames of targets according to the characteristic similarity. The method comprises the following steps that a tracking target is initialized, joint point information of all persons is counted, and invigilators are detected; matching a tracking target, carrying out similarity comparison on the characteristics proposed by the two frames of invigilators, and determiningthe tracking target if the similarity is maximum; and self-adaptive tracking: timely processing the situations of target loss, target wrong tracking and the like, and quickly and accurately finding alost target. By the adoption of the method, the position of the invigilator in the examination room can be accurately obtained, and reference is provided for subsequent action recognition of the invigilator.

Description

technical field [0001] The invention belongs to the technical field of object tracking based on computer vision and video comprehension, and more specifically, the invention relates to a posture-based invigilator tracking method. Background technique [0002] At present, after various examinations such as college entrance examination, postgraduate examination, self-examination and academic level test, a large amount of manpower is required to watch the examination video to analyze the examination style problems in the examination, such as inaction of the invigilator. When there are too many detection points, the staff is prone to visual fatigue, resulting in many false positives and missed negatives, and difficult video retrieval. The increase in personnel will cause a serious waste of human resources. Therefore, a big data analysis system for examination video or The method can automatically analyze the behavior of candidates and invigilators, and then analyze the problems ...

AI Technical Summary

Problems solved by technology

In terms of attitude tracking, Xiu Yuliang et al. designed an optimized network that establishes inter-frame relationships and forms attitude flows, and proposes a non-maximum suppression of attitude flows to reduce redundant attitude flows and re-establish temporally disjoint attitudes. The idea of ​​contact, this method has high tracking accuracy, but the problems of drift, occlusion and scale change in the long-term target tracking process have not been well solved

Images