Vehicle-to-Vehicle Distance Calculation And License Plate Recognition Based On On-car Video

As the important information source of intelligent vehicle, vehicle-mounted vision has increasingly wide range of application. Given the vehicle-to-vehicle distance calculation and license plate recognition based on vehicle-mounted video, the real-time and accurate vehicle-to-vehicle distance calculation can be applied to the early warning of vehicle collision, securing driving safety, the vehicle-to-vehicle distance calculation combined with floating car data(FCD) can be used to estimate traffic flow, determining the degree of road crowdedness, the license plate recognition based on vehicle-mounted video combines with the FCD can obtain abundant vehicle information which can be used to analyze cars driving track, and calculate cars density.Aiming at the applications above, this paper presents a method to calculate vehicle-to-vehicle distance and recognize license plate based on on-car video. The main work is as follows:Vehicles recognition based on on-car video. A well-trained an Adaboost classifier based on Haar-like feature is used for detecting vehicles. In the process of detection, the method detects preceding vehicles after copying every frame of video and reducing the its size, records the coordination and then the system amplifies vehicles’ coordinates and extracts the vehicles images in original image.License plate location based on vehicle image. In reality, the distance between the different preceding vehicles and the camera may vary, which leads to an apparent differences of the vehicle images’ feature. Given the difference of texture, Maximally Stable Extremal Regions(MSER) and color, this paper introduces three methods respectively, and in turn integrate them into a fusion method in order to locate license plates precisely.Vehicle-to-vehicle distance calculation based on license plate location. License plate has a fixed height, and the angle change of the vehicle does not affect the change of pixel height in on-car video. The calculation of vehicle-to-vehicle distance in this paper is according to camera image principle, using the pixel height of license plate and the fixed ratio between pixel height of license plate and vehicle-to-vehicle distance, finally acquiring linear distance on the basis of the position of vehicle in the image. Synthesizing the function of preceding vehicles detection, vehicle-to-vehicle distance detection, this paper presents the frame of vehicle collision avoidance system and floating car data collection system.Based on the result of license plate location, this paper introduces the algorithm of license plate recognition. 2 kinds of Back Propagation(BP) neural network trained by extracted eigenvalue of characters is used for recognizing license plate.