Real-Time Pedestrian Detection Based on Improved Gaussian Mixture Model

Applying image processing technologies to pedestrian detection has been a hot research topic in intelligent transportation systems (ITS). However, the existing video-based algorithms to extract background image may suffer their inefficiency in detecting slow or static pedestrians. To fill the gap, an improved Gaussian mixture model (GMM) for pedestrian detection is proposed in this paper. Three novel components have been incorporate into the traditional model. Firstly, the phase of graph segmentation is added before conventional parameters updating. Secondly, a mergence time adjustment scheme is employed to prevent foreground from merging into background. Thirdly, the notion of average weight is introduced as a secondary judgment criterion of foreground segmentation. To show the performance of the proposed method, this algorithm is applied into the real videos for pedestrian detection. The results show the accuracy and adaptability of this proposed method are over standard GMM.