The Research Of Crowding Detection Algorithm In Bus Based On Video Image Processing

Gaining the bus passenger information in real time is the key to the implementation of intelligent bus scheduling.Target at the problem that the way to get bus passenger information is outdated, this paper will get distribution image of passengers in the bus through the video image processing technology used to detected crowding on the foundation of front camera video surveillance, Thus improve the method of collecting passenger information, it is improtant to optimize bus scheduling and improve the quality of public transport service.First of all, given that the vehicle video surveillance can be influenced by vibration of the vehicle body and electromagnetic interference, this paper adopts the common image preprocessing methods: visually analyze image information distribution via Gray histogram,effectively extract target areas you are interested via gray-level threshold transformation and other methods, Second, aimed at characteristic in bus seat areas, take reflection shelter and other situations into account, manually select seat background areas, detecting passengers in front seats by integrating HSV color model algorithm with inter frame difference method.When all the front seats are occupied, the crowding detection in the passage areas will be carried out. Manually select three passage background areas. Respectively judge three areas in the passage areas via HSV color model algorithm and texture analysis method. The realization of judging different passage areas by ectracting color feature and effective texture feature parameters in the passage areas. District judgment can ensure that improve the accuracy of fusion algorithm while it adopts to realistic scenes. Bus crowding detection can be synthetically determined according to the judgment of each area. Finally, design the GUI interface. according to the operation of the experiment, the system should have the functions:Picture Reading, Data Input, etc. Aimed at practical function, the interface set with Open,Input, and other dialog boxes. The practicability of the algorithm is verified via the simulated interface.There are 3516 bus video surveillance images detected in the experiment, the overall accuracy is 93.6%, the experiment costs sixty-four minutes four seconds, the results further show that this algorithm has high accuracy and good real-time performance.