Specific Pedestrian Continuous Tracking Algorithms Research And Implementation Under Dynamic Scene

Pedestrian tracking is an important research direction in the field of machine vision. By analyzing the contents of the input video intelligently we can detect the pedestrians existing in the video. Further more, we can accomplish the pedestrian tracking task eventually through linking the track of the pedestrians.Specific pedestrian tracking refers to the process of the continuous tracking for the specified pedestrian, which has a broad application prospect in the field of the personal/family service robot. On the basis of the pedestrian detection algorithm presented in this paper, we put forward the specific continuous pedestrian tracking framework. Through the hierarchy classification, the pedestrian tracking task can be executed quickly and efficiently. The specific work is as follows:(1) First of all, this paper introduces the research background, the current researches for pedestrian detection and pedestrian tracking, analyzes the existing shortcomings. Then, some relative theories in this paper are introduced:support vector machine (SVM), Kalman filter and visual attention mechanism.(2) Aiming to solve the computational expensive for traditional pedestrian detection algorithm, an Edge Symmetrical pedestrian detection algorithm is proposed in this paper. Utilizing scan lines mechanism, the symmetrical value of each pixel is calculated along scan line and candidate regions are picked out. Then Candidate regions are verified by using Histograms of Oriented Gradients (HOG) feature and Support Vector Machine (SVM) classifier. Experimental results show that ESPD performs well in terms of accuracy and efficiency.(3) A target pool based specific pedestrian continuous tracking framework is proposed. Framework mainly divided into three levels:detection layer, evaluation layer and management layer. Detection layer detects existing pedestrian in input image, and submits the result to tracking layer for the follow-up tasks. Evaluation layer accomplishes tracking task preliminary, and submit the result to management layer for final decision. Management layer makes final decision according to the results. With the content in the target pool, management layer determine the results submitted by evaluation layer. Operate with the three layer, specific pedestrian tracking task can be completed. What’s more, this tracking algorithm introduces visual attention mechanism to improve the tracking speed and accuracy. Experimental results show that our ESPD algorithm is fast and robust for specific pedestrian tracking.