Research Of Target Vehicles Searching Based On Video Image And System Design

Traffic monitoring has been widely used today. This has been proved great convenient in traffic surveillance. Tracking vehicles involving criminal cases proved to be difficult in the past, such as traffic accident vehicles, vehicles been stolen. The intelligent search based on surveillance video of the video in this area will be a huge power. This paper is from this application, and study the algorithm of moving object segmentation in video, feature extraction, classification and recognition, as well as other related topics. Designed and developed a set of images based on video surveillance target vehicle search system.This thesis covers video image processing, pattern recognition and artificial intelligence. Some methods for digital image processing were discussed in this paper, focusing on segmentation algorithms, including background subtraction, histogram statistics, corrosion and expansion, blobs statistics. Motion object segmentation methods in video sequence image are varied. In this paper,, we use using the background subtraction method to segment video image, focusing on the background update model. Discussed background median filtering method, adaptive background model, Gaussian background model, and discussed the background update strategy. Propose the algorithms of HSI color model and shadow elimination algorithm, also the strategy of adaptive background updating. By extracting features of the vehicles in the video, the paper proposes different treatment from the vehicle style, color classes and license plate. All these have both connections and differences. Color recognition based on the HSI model was treated by quantify, which being improved and optimized. Vehicles recognition use the same methods of segmentation. Vehicle identification is based on the feature of the top length, roof height, front and rear ratios, These features were extracted using projection histogram method, and the selection of threshold proposed by adaptive threshold update algorithm, which is convenient and effective.Classification and Identification principles is another focus of the study,which were discussed 1 in the paper detail. The neural network BP algorithms are applied to all these classifiers, the classifier of the color, the classifier of the license plate character recognition, and of license plate recognition. Considering the complexity,we design three classifiers. In order to reduce the algorithm complexity of the character with a learning algorithm using the improved algorithm for momentum items, improve the efficiency of the system to learn.System implements modules of the image pre-processing, vehicle segmentation, vehicle identification, color recognition and license plate recognition. Each module has to opt for the realization of highly efficient algorithms and it is optimized for the goal of maximum efficiency in practice.