Research Of Vehicle Type Recognition Based On Video Sequence

As an important branch of Intelligent Traffic System (ITS), vehicle recognition has broad application prospect in detecting vehicle theft, regulating traffic order and automatic fare collection on highway, etc. Compared with other methods, vehicle recognition based on surveillance video does not require additional induction devices in the target area. Thus, costs are effectively lowered and surveillance video can provide more abundant vehicle information. Therefore, vehicle recognition algorithm based on surveillance video has achieved numerous research results.Based on the theories of moving vehicle detection, feature extraction of vehicle type and pattern recognition, this paper designed a vehicle type recognition system that integrated combined features and improved PSO-SVM. Main results are as follows:1. Shadow removal was carried out according to YCrCb color feature and LBP textural feature after the foreground image of moving vehicle was obtained by background differencing method based on mixed Gaussian background modeling.2. Regarding selection of features, it proposed that combined features of vehicle image, including Hu moment feature, LBPS feature, and length-width ratio feature, should be utilized to describe the feature information of vehicle, so that some problem, such as the susceptibility of single feature to lighting, weather and noises, etc. as well as the limited recognition accuracy, can all be solved effectively.3. A new LBPS feature was proposed. Based on LBP feature, spatial information was added to the texture information of vehicle as per the ideas of partitioning and information entropy. Meanwhile, dimensions of textural features were reduced.4. A modified particle swarm optimization (PSO) was proposed to carry out parameter optimization of SVM. Followings for random particles in the velocity updating formula were added. The concept of momentum was then introduced. By considering both the velocity changes at the former moment and the moment before the former moment, the problem that basic PSO was likely to fall into local extreme point and serious vibration at later period was solved.