The passenger flow information can assist to make more scientific departure plansand scheduling plans, which enables existing bus resources to be utilized better and getsthe best operational quality. So far, the common passenger flow statistical method can’tobtain precise result because of the huge passenger flow and disorder when passengersget on and get off.Since the result of monocular vision-based bus passenger flow statistical method isnot accurate, this paper proposed a video-based bus passenger flow statistical methodwith passengers’ multi-behavior analysis, in which the complexity and diversity ofpassenger’s behavior are considered.In view of the image sequence characteristics in bus environment, the targetdetection method is divided into two steps. In the first step, we primarily locate thepassenger area with four inter-frame difference multiplication algorithm. In the secondstep, we filtrate the passengers’ head using Hough Circle method and combining thegrayscale characteristics of the head.The passenger near the door whose behavior is with complexity and diversity willdisturb the counting criterion. On this basis, a new counting criterion is proposed. In thispaper we use trajectory clustering method to analyze the passengers’ behavior. Thencombining the spatial feature and orientation feature of the trajectory, the trajectorydistance is calculated. Finally, we analyze the effect on counting criterion of thebehavior which corresponds to clustering result.The experiment results show that the proposed video-based bus passenger flowstatistical method with passengers’ multi-behavior analysis can count the number ofpassenger more effectively. Moreover, comparing to other counting criterions, theproposed counting criterion is more stable and precise. |