Passenger Flow Statistic And Bus Lane Recognition Algorithms For Intelligent Bus System

With the sustained and rapid development of the economy of China, dramatic increasing citizens have resulted in a significant rise of family cars. Increasing threat of traffic jam and pollution as well as traffic accidents have seriously influenced the healthy development of modern cities, and city buses have become the best solution of this problem because of its high efficiency and environmental protection. By studying the development history of city transportation system, almost all cities around the world experienced the process of developing the city public transportation instead of developing private cars. However, most city public transportation system in China has the problem of improper distribution of resources, which result in passenger’s difficulty of travelling and transshipping and influenced the experience of citizens.Recently, the public transportation industries around the world are vigorously developing the intelligent bus system. One of the key function of the intelligent bus system is passenger flow statistic, because accurate and real-time passenger flow information will help the bus corporation arrange bus lines and stations more reasonable, and realize real-time scheduling according to current passenger flow information. In addition, the security of buses is also a hot topic. The intelligent bus system could help drivers control the operation of vehicles and improve security based on analyzing the road situation and roadside marks.For these cases, this thesis completed the following work:1. The thesis amplified the most commonly used depth estimation algorithms, edge detection algorithms and line detection algorithms, and select appropriate algorithms to improve and apply for the proposed passenger flow statistic system and bus lane recognition system.2. The thesis proposed a solution plan of passenger flow statistic based on stereo matching. After installing a pair of binocular cameras right above the bus door, some preprocess such as down sampling, calibration and decreasing contrast will be conducted on the collected binocular images, and depth map will be achieved by a depth estimation algorithm based on global matching. The core of the algorithm is graph cut algorithm. This thesis achieved a more accurate depth map by improving the matching criterion and the process of max-flow min-cut. On the basis of this depth map, the algorithm searches for those bigger disparity areas and confirm those areas which matches the passenger features as recognized passenger information. By tracing the moving track of passengers, the behavior of getting on/down the bus will be confirmed.3. The thesis proposed a solution plan of bus lane recognition based on Hough transform. Road images will be collected by cameras installed in front of the bus, then the image pixels will be transformed into HSV space and areas matched with bus lane features will be extracted by region grow algorithm. After edge detection, Hough transform will be conducted on the edge image to find out existing lines in the frames. Lines with reasonable angle or other features will be recognized as bus lane.According to the results of on-bus test, the proposed algorithm could achieve an identification accuracy of over 97 percent under the condition of non rush hour while 90 percent rate under the condition of rush hour. For bus lane recognition, this algorithm could achieve a satisfactory recognition rate under all kinds of environmental conditions.