The System Of Intelligent Visual Invigilating Video Based On Deep Learning

In recent years,with the continuous improvement and development of deep learning algorithms,especially in combination with target detection,breakthrough progress has been made.The third generation of intelligent visual monitoring technology combined with computer vision technology has also achieved leapfrog development.Therefore,in view of the social problems caused by increasingly cheating on the examination,intelligent visual monitoring system based on deep learning has great significance and value in eliminating cheating in exams.This paper aims at achieve a set of detection system for cheating in monitoring video under the scene of examination by using the study of human behavior recognition and object detection algorithm,and using the object detection algorithm based on deep learning,which can achieve real-time detection,cheating of alarm and the management of detection results.In view of the above goal,this paper has done the following work:(1)Research on human behavior recognition and object detection algorithm.This section includes a study of existing technologies for human behavior recognition and the current algorithms for object detection.This paper focuses on the application of deep learning algorithm in the object detection algorithm based on region proposal and regression-based target detection algorithm,and compares the performance of YOLO and SSD algorithms to detect cheating behavior under the Caffe framework.Therefore,we found the suitable core algorithm for the detection of cheating in the examination room scenario.(2)According to the uniqueness of this topic,we set up an image dataset of cheating in the examination room scene.In the dataset,human behavior(cheating and normal behavior)in two kinds of examination rooms were manually labeled,and three specific types of cheating behaviors((Passing the answer,Peeping at others,Peeking the mobile phone),so the number of original images in the data set exceeds 10,000 and the number of annotated samples exceeds 40,000.This dataset can be used to study the algorithms of cheating behavior matching and can also be used to augment the basic research data of human behavior recognition.(3)Realization of intelligent visual monitoring system.After comprehensively analyzing the performance of different algorithms,we chose SSD as the core detection algorithm and improved the SSD detection accuracy by incorporating metric learning.Finally,a solution of intelligent visual monitoring system based on deep learning is designed:the scheme of dual-modules cooperation is adopted to achieve the good effect of video monitoring and management.This paper describes the system principle and implementation process,and then achieves the expected result through the realization of a complete system on the Windows to satisfy the real-time and operability.