Human Object Detecting Based On Classroom Recording System

With the rapid development of modern society and economy, video surveillance systems based on video content analysis have an important impact on security, education, artificial intelligence and other industry. Recording System, as an assisted teaching system for education, plays a increasingly important role. However, the traditional recording system is still human controlled.As a classical topic of intelligent video analysis, visual target detection and tracking has also been rapid developed. Based on target detection and tracking algorithm, the recording system can automate the task of class recording control by analyzing jointly visual target detection with the help of multi-cameras, which provides intelligent and convenient recording services.This paper summarizes the existing pedestrian target detection algorithms and achieves an object detection algorithm under multiple camera surveillance based on the combination of inter-frame and background difference in student detection module of recording system. What’s more, a CENTRIST pedestrian detection method based on edge classification is proposed in teacher detection module. We introduce the algorithms into application to fulfill the demand of student and teacher locating module in recording system.For student detection module under static background, considering the high complexity of optical flow algorithm, instability of the inter-frame difference methods and low robustness of background subtraction, this paper achieves a background-modeling algorithm under multiple camera surveillance based on inter-frame difference. By updating the dynamic background, the adaptive background model enhances the robustness with the same efficiency of inter-frame difference methods.For teacher detection module under dynamic background, this paper compares the existing target detection characterization such as HOG, CENTERIST. To overcome the noise immunity of CENTRIST detection characterization, we propose a pedestrian detection algorithm based on edge scale selection. Since the outline information plays an important role in pedestrian detection, the proposed method extracts major outer contour by edge grade and removes the local texture in environment and internal area of target to improve the performance and robustness of SVM classifiers. Moreover, the pedestrian detection based on motion region reduces the missing and false detection in practical application scenarios.Finally, with the application demand of class recording system, this paper proposes the system design, software and hardware architecture based on TMS320DM8168 embedded platform and Linux server. Also we validate and optimize the student and teacher detection algorithm in practical embedded platform, and set up multi-core collaboration, client-host operating mode with Linux server as logic control core to achieve the automatically class recording system.