Vehicle Tracking At Night And Evaluation Technology

Vehicle detection and tracking technology is an important content in the field of intelligent transportation. It is the primary goal of the "the five-years plan" to constructe the wisdom city. And it also provides key technonogies to constructe the safe city. Moving targets detection and multiple targets tracking technonoly has been a research focus in the areas of computer vision and machine learning. In this paper, we put forward a method, which based on automative lighting characteristics to detect and track vehicles at night, on the basis of previous research work. And the method training classifier to reduce the false rate and track vehicle robustly at night. Evaluation technology evaluate the vehicle tracking. And putting forward some improvements on performance and effieciency.The main contribution of this thesis are as follows:1. Reseraching vehicle detection based on the characteristics of lamp light. At first, this paper compares the variety of vehicle detection and tracking technologys, for example interframe difference method, background difference method and adaptive threshold segmentation method. The aforementioned methods can be efficiently applied to daytime traffic scenes with stationary and unchanged lighting conditions. However, in poorly illuminated or nighttime conditions, these methods may be unreliable. In the training phase, using the Haar features and mulit-scale genmetric, shape feature to train the Adaboost classifier. In the testing phase, according to the histogram statistics sets the threshold to segment image at night. All detected light through the Adaboost classifier classification. According to the characteristics, such as speed, geometry, shape similarity of data association to establish the mulitiple target tracking.2. Evaluation technique for vehicle tracking. Evaluating the vehicle tracking is a huge workload, therefore, this paper puts forward to a method to automatic evaluate the properties of vehicle tracking. A semi-automatic annotation tool is put forward to get the actual vehicle tracking. Comparing the Ground Truth and Detected Results of vehicle tracking by evaluation algorithm. Getting performance value, such as ture rate, miss rate, false rate and switches. According to the result, the method in this paper can carry on the tracking vehicle robustly.