A Method Of Bus Passengers Detection Based On Video

Accurate and comprehensive bus passenger information is the basis for publictransportation scheduling and planning. Compared to the traditional optical passengerdetection methods, video detection method not only can make full use of the videosurveillance systems, but also has many advantages such as wide range of signal detection,complete target information, etc. This method can successfully achieve passenger detectionunder the crowded and disordered situation, so, it’s more suitable for the bus environment ofour country. This paper focuses on the passenger detection technology based on video, and afast, accurate and efficient passenger detection and tracking algorithms is proposed.Firstly, according to the background and foreground characteristics in image sequence,proposed a region of interest detection method based on frame difference, subtracts twoadjacent image and filter the results with connected domain to get the target region of interestwhere contains moving passengers. This process is not sensitive to light, and also caneliminate tiny changes of the background which caused by the vehicle’s jitter, has a strongadaptability.Then, in the passenger area of interest, a feature-based passenger’s head extractionmethod is proposed. In this method, gray stretch, enhanced the head target identification,overcome the impact of head shadow and light insufficient; morphological fix process,eliminated the tiny fragments and slim connected component interference, fixed head outlinein binary image; by the head outline similarity objective function, achieved accurateextraction for the head target. This head extraction method is simple, accurate and real-time.Finally, track the detected passengers target by Mean-shift algorithm, can overcome thehead target partial occlusion, outline and size changing during the movement. Besides, thepaper established the association between tracking and detection, so that the trackingalgorithm can be closely combined with the detection algorithm to achieve accuratelypassengers counting and direction determination.Large number of simulation experiments show that the method proposed can effectivelydetect passenger information with good accuracy, timeliness and noise immunity.