SSD-based Smart Classroom Multi-category Person Detection And Recognition Under The Caffe Framework

Multi-category character detection and recognition in wisdom classroom is great significant in education.The research this technology was proposed and developed widely.It can let us know whether the students listen to the class carefully and the specific status in real time.The feedback of the distribution trend of each student's concentration in a class were displayed.We can make a targeted course arrangement to improve the teaching quality.Student's specific class status of in school and the quality of teachers and other information can be acquired.So this technology plays a certain role in promoting the progress of education.The paper was based on deep learning,the related technology of human posture detection and recognition in classroom scene.SSD algorithm was proposed to detect and recognize classroom characters.The whole detection and recognition process is divided into three steps: character coordinate position regression,attitude classification and engineering design.The real-time detection of the character position can be realized and the character class status in the complex classroom scene can be recognized.In this paper,according to the detection stage,the characteristics of SSD algorithm are analyzed,the multi-layer convolution features are extracted by SSD,the multi-scale feature graphs are extracted from the skeleton network,and the candidate boxes with different number and aspect ratio are designed based on the feature graphs.Detect the target people in different scales;In the end,the location of each target is predicted,the detection accuracy is 98.9%,and the recall rate is 91.9%.In the attitude classification stage,the Res Net network structure is used to extract features,combined with the actual use requirements,design different network structure,output channel number,convolution cascade,extract different depth target features to judge,and realize the classification of multi-posture of characters.The accuracy of classification was 99.2%.The classroom information...