Researches On Vehicle Detection And License Plate Recognition Key Technologies Based On Video

Vehicle detection and plate recognition play an important role in intelligent transportation system. This thesis concentrated on several key technologies which used to deal with the problems existed in vehicle detection and plate recognition. Those technologies can improve the vehicle detection and plate recognition performance effectively. The main work includes following aspects:1. An improved frame average method was used to initialize background. Firstly, inter-frame difference method was used to choose some frames with small traffic flow, and then pixels detected as background are being used to initialize background. In the process of background updating, only the pixels which belong to background are being used for updating. This algorithm can improve the prcision of background initialization and the speed of background updating.2. Several commonly used vehicle detection methods were studied. An effective vehicle detection method which fused three inter-frame difference, background subtraction and edge-based background subtraction methods together was proposed. Firstly, three inter-frame difference, background subtraction and edge-based background subtraction methods were used respectively to obtain three foreground images. Then, D-S evidence theory was introduced to fuse those three foreground images to get the final vehicle regions. This algorithm makes full use of information of the three methods, improved the performance of vehicle detection.3. Algorithms which have proposed in literatures were studied. An effective license plate location algorithm based on edge and Adaboost was proposed. In the first step, candidate regions of license plate are extracted by edge statistical analysis. Then, a cascade classifier based on Haar-like features is constructed to reject non-plate regions and obtain plate region. This algorithm can take full advantage of the method based on edge and Adaboost in speed and low false alarm respectively, which dramatically reduces the computational complexity while maintaining high detection accuracy.4. A plate character segmentation method based on template was proposed. This approach was based on Chinese license plate priori knowledge that 7 plate characters have the same width and height, with character aspect ratio is fixed. This method segments all characters successfully even when there are broken or connected characters while maintaining high speed.5. Several character recognition methods such as template matching, neural network and SVM are studied. An efficient character recognition method which fuse template matching and SVM and takes full advantage of those two methods is proposed. The experiments results show that this method can improve the character recognition rate by dealing with the recognition problem of degraded character and ambiguous character.6. Software for vehicle detection and license plate recognition is realized by using OpenCV and Visual Studio 2008.