Color Video Based Multiple Passengers Moving Object Recognition And Application On Bus

Automatic Passenger Counting technology is one of the most effective ways to collect the time and location data of passengers boarding and alighting at different stops. And it is the foundation of transit service planning, scheduling, and forecasting. Because of the high density of passengers flow and crowded aboard and alighting situation, the doorway infrared and treadle mats APC devices which are widely used abroad perform not well and fail to meet the demands of passengers data collection in our country.Multiple passenger object recognition is the central technology of APCs. It directly influences the accuracy of APCs. In order to solve current drawbacks of APCs, this thesis proposes a new multiply passengers object recognition algorithm which based on the video image processing platform. In this case, the system inputs the real-time images from the camera, and outputs the number, location, area of human head.In this thesis, a multiple passengers object recognition method is introduced which includes multiply passengers detection and multiply passengers tracking. According to the feature of passengers in the images, this thesis applies an object detection algorithm which combines the gradient algorithm and histogram algorithm. The color histogram algorithm performs well when the object border changes greatly while the gradient algorithm performs well when the object background interrupts the object detection. Therefore, according to combine these two algorithms, the objects detection could have a good accuracy. The algorithm in this thesis firstly applies gradient information for diminishing the interference introduced by the background and then applies the color histogram algorithm for objects detection. What's more, this thesis introduced meanshift tracking algorithm to track the detected objects. Because the meanshift algorithm is based on the color information and non parameter probability model, it is robust especially when the shape of the objects changed.In this thesis, a lot of experiments and large mount of images are used to test the proposed algorithm. The multiple object recognition based APCs could realize to count the boarding and alighting activities and the accuracy could be around 90%. The results show that the proposed algorithm could have a good performance on multiply passengers object recognition. Compared with other algorithms, the proposed algorithm could have a better accuracy.