Research On Classroom Behavior Intelligent Image Recognition And Analysis

This topic conducts research on classroom behavior intelligent image recognition analysis.Combining intelligent image analysis with educational scenes not only achieves higher accuracy of behavioral action recognition in the classroom scene,but also the time efficiency of the algorithm is higher.According to the process of behavioral action classification and recognition,the system is divided into three steps:video-based moving target positioning,human behavioral action feature extraction,and human behavioral motion recognition classification.With regard to video-based moving target localization,this paper proposes a fast moving target foreground detection algorithm based on FFmpeg codec.The time efficiency of foreground object detection using the commonly used code table method,mixed Gaussian model,Vibe method,etc.is too low,and the single frame image takes about 100-200 ms.For the H.264 encoded compressed video,the FFmpeg codec is used to process and analyze the video stream,and the macroblock type table and the motion vector table can be easily extracted.Then,according to the information,the motion vector map is restored,and finally the video image preprocessing related knowledge is used to perform image filtering and morphological processing to realize the detection and localization of the moving target foreground.The algorithm can effectively locate the moving target foreground area in 10ms,and the time efficiency is greatly improved.In this paper,an improved MHI-HOG joint feature is proposed in this paper.The accuracy of behavioral motion recognition based on the HOG feature of a single frame image is relatively low.The motion history map contains the timing information on the motion compared to the motion image of the single frame.Therefore,the HOG feature of the extracted image motion history map is more representative of the human behavior action than the HOG feature of the directly extracted image.In this system,by using the fast moving target foreground detection algorithm based on FFmpeg codec,the motion foreground map of the video sequence frame can be quickly extracted,the motion history map is continuously updated according to the image,and the HOG feature of the motion history map is finally extracted as The characteristics of human behavior.It is verified by experiments that the accuracy of the joint feature is significantly improved in the recognition of human behavioral motion compared to the directly extracted image HOG feature with almost no increase in time overhead.For the classification of human behavioral motion recognition,this paper proposes a fast neural network-support vector machine joint recognition classification method based on lookup table.The scheme fully exploits the advantages of neural network structure in multi-classification recognition and support vector machine in two-category recognition.By training the neural network,it can effectively classify and recognize human behavior.When the neural network discriminates that the probability of an action does not exceed 60%,based on the way of checking the table,the appropriate support vector machine is selected for joint determination to complete fast and effective classification and recognition.The principle of the support vector machine check table is to use the neural network to determine the top two SVM classifiers of the classification results.Based on the MHI-HOG feature,the behavioral recognition rate of the Weizmann Dataset in the public video set reached 97%,which was significantly improved compared to the 90%recognition rate using the HOG feature and the SVM classifier.Finally,this paper builds a Hi3531A-based classroom behavior intelligent image recognition and analysis system.The classroom intelligent image recognition analysis software is cross-compiled under Ubuntu 15.04.Real-time recognition and analysis of classroom behavior is achieved through modular operation and multi-threaded operations.Using the software and hardware platform used in this topic,the experiment of behavioral action recognition analysis in the classroom scene was carried out.The experimental results verified that the system achieved the expected performance.