Head Detection and Tracking in Top-View Videos Using Circle Hough Transforms

• Main Authors: Po-yi Lee, 李柏儀
• Other Authors: Tang-kai Yin
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/60885121511128430782

碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 103 === In top-view surveillance videos, head tracking can monitor the number of persons entering and leaving a building without more hardware requirements. In this research, we tried to use circle Hough transforms (CHT) to do head tracking. The CHT is one of major tools to find round objects in images with many successful applications. However, human heads are not pure round objects, but have variations such as different hair styles, bald, and wearing hats. To overcome this, repeated tests were made on the same image objects across frames. From binomial distributions, the increased recognition rates were mathematically validated. In the experiments on the videos without nonhuman objects such as umbrellas, the CHT with repeated tests have over than 90% accuracy. Finally, experimental results showed that in crowded top-view surveillance videos, our system can effectively track and count the numbers of people ingoing or outgoing the gate with the average accuracy rate of 88.68%.