Passenger Flow Detection In Rail Transit Based On Haar Features

In recent years, along with the rapid development of our country’s economy, the urban rail transit has become a main way for people to travel. How to monitor people’s activities effectively to ensure the security of the passenger is particularly important in the rail transit operations. In passenger security monitoring, counting the number and estimating the movement direction of the passenger flow are two key technical problems. According to the video information shot in the subway station, we count the number and analyze the movement direction of passenger flow. The acquired results can provide a theory basis for the related departments to deal with emergencies and make decisions. The main contributions are described as follows:(1) Aiming at the distinct differences between front heads and back heads, a passenger flow moving detection method is proposed in two opposite directions by using two trained classifiers simutaneously. Based on P.Viola haar feature set and Adaboost algorithm in Open CV, the two samples of front and back heads are trained to realize the head detection respectively in the classifier. The proposed method is compared with a single classifier. The experiment results show that the detection accuracy is enhenced, and the performance of passenger flow detection system is impoved.(2) According to this drawback that color histogram fails to describe the spatial information and only provides the frequency of diffenrent colors appearing in the image, an improved human matching method is proposed. The part of the human body from the head to the shoulders are intercepted to create a HSV color histogram, which is used for matching and recognizing of the characters in adjacent frames. In order to compensate the lost spatial information, an effective measure is utilized to predict the location of the characters in the next frame, and the seted threshold, for the subtraction result of the pixel value of the location coordinates in consecutive frames, is distinguished again. The proposed method is comlpared with the tranditional human matching method. The experiment results show that the matching accuracy is increased.(3) Based upon human’s color histogram, an improved Camshift algorithm is proposed. By comparing the distance of human’s color histogram directly, each frame is processed. Then the people’s histogram and its location, speed and other information are updated constantly, and hence a human characters tracking information chain is created effectively, which can realize continuous frame trajectory tracking. By comparing the location coordinates in successive frames, the moving direction of the passenger flow can be predicted. The experiment results show that the method is effective.(4) A passenger flow detection system is developed based on C/S framework. The system has analyzing, detecting, recognizing, tracking and counting functions, and provide a theory and actual basis for the management and decision-making of emergency in daily operation of the rail transit. The results of the test indicate that the system has strong robustness and real-time.