Summary: | 碩士 === 國立高雄應用科技大學 === 電子與資訊工程研究所碩士班 === 94 === The statistics of pedestrian movement which to be obtained from particular entrances and occasions are highly consultable. However, the manual counting method takes highly manpower costs. The rotating axles are not easy to be assembled and only one person can pass it at a time. By using electronic mechanisms, such as infrared/ultrared rays; the problem regarding number of persons still remaining unsolved. Due to the above reasons, we propose a novel automatic people-counter approach based on video processing technique. The cost of this mechanism is more inexpensive and easy to be assembled. Furthermore, the problem regarding high density of people can be solved by using this mechanism.
This thesis proposed a bi-directional people counting method which can be applied to situation such as high people-flow and can be utilized in real-time applications. First of all, the passing people are roughly counted with the area of people projected on an image captured by a zenithal video camera. The moving direction of the pedestrian can be recognized by tracking pattern of each person by analyzing its HSI histogram. Besides, the occlusion problem of both people touching together and merge/split phenomenon can be overcome in this system.
Finally, the experimental result shows that the accuracy is about 95%, and this method can meet the requirement of Real-time.
|