Researching The Method Of Low-resolution Vehicle Detection And Tracking Fast Crowded Scenes

Because the traffic condition is worse gradually, intelligent transportation systems(ITS) have been widely attention. Vehicle detection and tracking is the key and basic technology in ITS, and has become the hot study in ITS. At present, because the effect of existing vehicle detection and tracking algorithms can’t meet the practical requirement under the condition of crowded scenes and low resolution, they need to be further improved. To this situation, the contribution of this thesis is mainly as follows.1. A low-resolution vehicle detection algorithm is proposed based on the cascade Adaboost classifier. The framework of hypothesis generation and hypothesis verification is used to detect vehicles. In the phase of hypothesis generation, When the expectation and variance calculated within child window meet a certain condition, the cascade Adaboost classifier is used for validation. Haar-like features, as the input of the classifiers, reflect the texture information of the target, and are not sensitive to light. Moreover, in the process of training the cascade Adaboost classifier, low-resolution pictures of vehicles are selected as positive samples; and when child windows scan the detecting image, this thesis adopts the window pyramid method. A large number of the test results show that the algorithm has got very good detection effect under the condition of crowded scenes and low resolution.2. A real-time tracking algorithm is also proposed based on the vehicle detection. According to the color weight histogram in the vehicle area, the Euclidean distance between the vehicles in the previous frame and the detection result in the current frame is calculated, and when this distance satisfies a certain condition, the two vehicles in the two frames are considered as the same vehicle. The vehicles in the previous frame, which matched successfully in current frame, are estimated the position in the current frame by Mean Shift tracking algorithm. At the end, the vehicles in the current frame include the result of the detection and tracking. This improves the performance of detection in crowded scene, and also improves the tracking speed.3. The vehicle detection and tracking in the crowded scene and low resolution is applied to actual systems, and an E-traffic system has been developed successfully. According to the field test, all the performances can meet the practical requirement.