|
With industrial development and modernization of car traffic on the environmental impact of urban transportation, intelligent transportation systems (ITS) urban modernization has become an important area of ??research at home and abroad. Vehicle identification based on image processing systems, including vehicle identification, license plate recognition, computer vision and pattern recognition technology in the field of intelligent transportation applications, an important issue, the application range is very wide, and has broad prospects and great economic value. The main subject of an image processing-based vehicle detection system. Mainly from the two major approach: models of the recognition algorithm and traffic information of the video image acquisition system. The system focuses on the key technology of theoretical research and implementation, highlighting the characteristics of real-time and accuracy, we propose a modified moving object segmentation algorithm and on the specific vehicle recognition algorithms. The main work and innovations include: 1) presents a complex background environment of the moving target detection algorithm. The use of the characteristics of color images for background extraction, which effectively suppressed the vehicle with a similar background and light gray and other difficulties; against two different situations, namely put forward the corresponding background updating scheme; utilize binary image, we proposed a fast and efficient region filling algorithm, in order to meet the real-time extraction of moving targets. 2) design a set of feasible characteristics of the vehicle silhouette vehicle classification model, a classification scheme step by step, complete vehicle identification algorithms. Using the image information, proposed the concept of soft sensor, instead of the traditional hardware sensing devices; using Douglas-Peucker algorithm feature vector silhouette compress the information, reducing the complexity of the operations of data; using a two-step classification program, full use of the invariant moments, area ratio, length and other important information, obtain a good classification results. 3) implements a multi-channel video capture and compression storage systems, and on this basis, build a complete vehicle detection system implementation. Software uses a multi-threaded design, in order to avoid the problem of excessive resource consumption; and using third-party compression module, video capture and compression storage systems, good to meet the needs of practical application. This is characterized by: a complete implementation of a system can be applied AVI; propose a background updating algorithm; designed an edge contour features based on vehicle classification. Stable operation of the system can be installed with Microsoft Windows XP Professional (SP2) under a PC, PC machine CPU is P4 2.8GHz, memory DDR-512M. Used for experiments in two segments 25 minutes video of the test, the experimental vehicle classification accuracy rate above 90% per frame time-consuming less than 40ms, CPU occupancy rate of less than 40% of the full realization of the frame operation.
|