| According to the research of the technology of pedestrian detection and tracking in the intelligent video surveillance system. We propose a novel design for intelligently monitoring the detection of moving objects, pedestrian recognition, tracking and counting. The research mainly talks about the following items:1. We analyze three approaches of moving objects detection and according to the practical system we propose an approach of detecting objects depending on the concept of background difference method based on Gaussian mixture model.2. We studied the detection of moving objects. Then we proposed the feature fusion approach based on HOG-WLD feature which is studied upon the traditional single-feature pedestrian recognition. After describing the comprehensive description of the features of targets, we take it into the trained SVM classifier for enhancing the precision of the pedestrian recognition. For traditionally depending on the pedestrian’s shaped features to screen roughly, we introduce the approach that selecting roughly through edged symmetrical feature(physical feature)firstly, then precisely recognizing pedestrian through the HOG-WLD feature(abstract feature)gained by the feature fusion technology, thus the system provides the advantages of the reliability and effectiveness for enhancing the degree of accuracy.3. we have studied so many researches about pedestrian tracking that we conclude the main ideas about it. Then we implement Kalman filter algorithm to track pedestrians. In order to track multi-objects, we draw multiple interested goals by manual at first, then allocating the trackers automatically.4. We apply pedestrian detection and tracking technology in counting. For pedestrian counting, the system gives accurate counting for pedestrian coming from both sides into certain area. Experiments indicate that the system has a high degree of accuracy, what’s more, it can operate real-timely and robustly. |
PDF Full Text Request