Abnormal Behavior Detection Based On Head Movement Analyze In Examination Room

With the continuous development of computer vision and pattern recognition and the reduction of the imaging system cost,the intelligent examination system of the examination room will automatically recognize the face and locate the facial features to estimate the head movement attitude.According to the attitude parameters,the abnormal behavior analysis can effectively alleviate the censorship pressure.This paper mainly designs 2D motion estimator and 3D head motion estimator to realize intelligent monitoring,including image preprocessing,skin color extraction,2D motion estimation and classification,face detection and feature extraction and two-dimensional image human head motion attitude estimation.Based on the Multi-Block LBP feature-based AdaBoost cascade classifier training on MIT face map and monitoring system map,the classifier is loaded and real-time face detection is carried out.The face area is based on cascade convolutional neural network algorithm.Finally,based on the facial feature points,the attitude parameters(Euler angles)of the head are calculated,and the abnormal behavior is determined by horizontal rotation of the yaw angle.This process is the 3D head motion estimator designed in this paper.In this paper,we analyze the abnormal motion of 2D video frame with unsuccessful face detection and feature localization based on sparse optical flow,K-means motion vector classification and histogram similarity.The experimental results show that the performance of AdaBoost classifier optimized by this paper is validated by ROC curve,and finally the head motion attitude estimation is converted to perspective N point problem(PNP)through the precise coordinates of face feature points in DCNN neural network.The Yaw parameter can be used as the threshold for abnormality judgment by comparing the pose parameters of abnormal behavior frames and non-abnormal behavior frames in video memory.The experiment of 2D motion estimator designed in this paper shows that histogram comparison of the motion vector to the largest corresponding cluster center can be done.The bigger the motion vector is,the smaller the histogram comparison value is,and the abnormal threshold can be set by setting the appropriate threshold Behavioral decisions.In the final examination room,the 2D motion estimator designed in this paper performs face detection and feature extraction on each frame of the monitoring video to determine the abnormal behavior.The remaining video frames pass the calculated Euler angles of the head motion pose parameters Parameter Yaw parameter to determine abnormal behavior,by setting the appropriate range of Yaw values,you can distinguish a simple twist to cheat behavior,the algorithm can be based on the head posture parameters to complete the examination of abnormal behavior within the examination room.