Research On People Counting Of Public Indoor Places

Today's society, the rapid development of computer technology makes computing power improved largely, in the meantime, using computer to realize human's visual function has become one of the most popular topics of current computer field. As a typical application of computer visual field, people counting of public indoor places has important applications in many aspects such as statistical analysis of business data flow of people, Public Safety etc., it is actually mainly related to pedestrian detection and tracking, which are two hot research focuses of the field. However, the research on people counting of public indoor places has just started, accuracy and real-time of are need to be improved. Therefore, this thesis focuses on researching and designing algorithms of pedestrian detection and tracking that meet real-time and accuracy simultaneously to improve real-time and accuracy of methods of people counting of public indoor places, specific work started on the following aspects:First, research common algorithms of pedestrian detection and propose an improved algorithm based on that.In this thesis, some common algorithms of pedestrian detection were studied, an algorithm of pedestrian detection based on HOG feature and motion information was proposed, which combines static characteristics and motion information of target, and can meet real-time and accuracy requirements.Second, research common algorithms of pedestrian tracking and propose an improved algorithm .Kalman filter, particle filter and some other commonly used pedestrian tracking algorithms as well as the algorithm of multi-target tracking data fusion were studied, a tracking algorithm of multi-feature matching that is based on region was proposed. For block, stationary targets and other problems of multi-body tracking under complex background, corresponding solutions were provided.Third, design and develop the prototype system of people counting.On the basis of the above research, a system of people counting of public indoor places was designed and developed, which takes Microsoft Visual Studio 2005 as developing tool and uses the language of C++ as developing language. By analysing and processing real-time scene video, the system can detect, track and count pedestrian of public indoor places effectively, and finally store the results to the local database for further analysis and processing.Experimental results show the accuracy and validity of the proposed algorithms of pedestrian detection, tracking and counting.