| Summary: | 碩士 === 國立勤益科技大學 === 電機工程系 === 104 === The average daily customer flow is an important message for a public transit system or a service industry company, and the support manages make corresponding adjustments of strategies. A people counting system is usually positioned in the doorway or a corridor of a building. Turnstiles, optical sensors and image processing are often used in people counting systems. The image processing based systems for people counting have the problems such as image occlusion, shadow or mistake goods as people. In this paper, we propose a bidirectional people-flow counting system based on stereo vision. Two top view cameras capture images. The targets are detected by using the background subtraction method and the stereo vision algorithm with multi virtual detection lines. That solves the image occlusion and shadow problems and improves the system process speed. The sum of the absolute difference algorithm calculates the target height. The human height and the shoulders are as human characters to recognize a person from detected targets. That solves the problem for system mistake goods as people and improves the recognition system robustness. Finally, the people tracks are detected by template matching. Then the people number is counted when people pass through the count area. The people counting system effectively reduces people image obstructions and counts people number with high accuracy.
|
|---|
