Research Of Vehicle Detection Method Based On Features Fusion

In recent years, with the rapid development of transport and the rapid increasing of the number of cars, the road safety issues become more prominent, and in order to solve this problem, the researchers are working on various transportation aid systems to improve the driving safety through the Intelligent Transportation Systems. Vehicle detection is one important part intelligent transportation system; we may warn of the accidents that may occur by analyzing the driving environment effectively which can reduce the traffic accidents effectively.Nowadays, the feature-based vehicle detection methods were widely used by researchers. This method can be divided two parts:feature extraction and classifier training. Firstly, selected feature must have the abilities of characterizing the vehicle information. Multi-level vertical oriented gradients and multi-level local binary patterns can be a better description of the vehicle. Secondly, the selected classifier must have abilities of high classification speed and high recognition rate. Histogram intersection kernel support vector machine classifier has these advantages. According to above analysis, this thesis presented a vehicle detection algorithm based on features fusion and intersection kernel SVM.In this thesis, a method is designed which combined Multi-level vertical orientation gradient features and the Multi-level local binary pattern features as vehicle features. The principal component analysis is used in vehicle detection system in order to avoid the time consuming in the training model because of the high features dimension. The principal component analysis is used to decrease the dimension of the Multi-level vertical orientation gradient features, then combined the Multi-level local binary pattern features as the result feature. The support vector machine with the histogram intersection kernel is used for the features classification which can effectively shorten the training model and classification time.For varies scales of the pictures to be detected, zoom the detection window depending on the picture, and then slide the detection window to traverse the entire picture. This method may cause multiple test results of one object; we fused the detecting results to get the final test results with window fusion algorithm.To verify the performance of the algorithm, we carried out a comparative experimental and the results show that the vehicle detection features fusion algorithm in this paper can improve the accuracy of vehicle detection and reduce the false detection rate effectively, the vehicle detection performance improved from the overall. Finally, we summarized the work in this paper and also prospected the future work.