Research And Implementation On Traffic Monitoring System Besed On Smart Mobile Terminal

With the national vehicle’s rapid development,urban traffic problem becomes increasingly serious.Traffic monitoring is also put forward higher requirements.On the other hand, With the development of on-board monitoring technology and the popularization of mobile intelligent terminal make building a whole vehicle monitoring system based on mobile intelligent terminal possible.Among them, vehicle identification as the core of the vehicle video monitor system is also the key difficulty, has attracted scholars engaged in research at home and abroad.The current mainstream recognition algorithm based on method of background difference, frame difference method, optical flow method and the vehicle identification algorithm based on learning. These algorithms have their own advantages and disadvantages, suitable for different recognition scene.We want to be able to recording in the process of driving within video analysis and detection of vehicles on the road.Due to the weather and light, in the process of driving environment is more complicated.At the same time, the vehicle appearance is also relatively have obvious difference, Unfavorable apply background difference method and frame difference method. Optical flow method due to its high complexity, need large hardware cost, does not apply to ordinary traffic monitoring system.According to what we’re testing scenarios,after analysis we decide to handle vehicle identification based on learning algorithm.On the basis of domestic and foreign relevant research work, we finally chose two more mainstream in the field of vehicle identification algorithms:Hog+SVM and Haar-like+AdaB- oost,Among them,Hog+ SVM algorithm is more accurate but slower than Haar-like+AdaBoost. According to the characteristics of on-board monitoring video we found.Vehicles, the size of the car is changing when driving.The distance of the vehicle in the picture is displayed in a smaller, while nearby vehicles display is bigger, so you must choose a different size to identify areas, at the same time, considering the distance is too small vehicles need to have a higher recognition accuracy of the detection method.Based on the above analysis, the final decision will Hog+SVM and Haar+ AdaBoost combined two kinds of algorithm, according to different object recognition play their respective advantages.In the algorithm optimization, we mainly based on Kalman algorithm to predict areas to be detected, jump frame processing at the same time. In does not affect the detection accuracy, further optimizing the performance of algorithm, and improve the system of the user experience.In addition,We encapsulate the above algorithm,we implements a relatively complete monitoring system intelligent mobile terminal based on the Android system, including the real-time processing of video streams, the recognition algorithm of encapsulation and call lookup, video playback function, and recently a gasoline station. In addition, the data storage system for mobile, also do some optimization.Finally, we in the mainstream of the Android platform experiment, the experimental results show that the 10 times the average processing time is 121.679 milliseconds, basic can meet the needs of processing speed. Part five times the experimental results show that the detection rate of 90.41%, full detection rate of 34.25%.Algorithm accuracy completely, need more training samples.