Abstract: According to the characteristics of the traffic environment at night,a method for nighttime motion vehicle detection based on self-adaptive character selection is presented in this paper.Firstly SIFT algorithm is used to extract the shape characteristic of the motion vehicle at night and obtain the feature vector by the integration of color and texture feature.And then Boosting algorithm and Genetic algorithm are used to obtain a set of fuzzy rules and regulations corresponding weights in the form of iterations and achieve self-adaptive feature selection by weighted voting decision to the most favorable characteristics of the detector.Finally the experimental results show that this method can deal well with complex situations at night,such as occlusion,illumination and background interference,maintain real-time vehicle detection in the choice of different self-adaptive characteristics to background information,and achieve the real-time and accuracy requirements of traffic intelligent monitoring system with strong robustness,which are tested in three different nighttime traffic scenarios.

Key words: vehicle detection, SIFT character, Boosting method, self-adaptive character choice