Vehicle ROI Extraction Based on Area Estimation Gaussian Mixture Model
ZhaoNan Huang, Huabiao Qin, Qing Liu · 2017
The extraction of vehicle region of interest (ROI) is a key step of vehicle detection system. When Gaussian Mixture Model (GMM) is applied to vehicle ROI extraction, the extraction speed is unsatisfied and the results remain some noise blocks. In order to solve these problems, a method of vehicle ROI extraction based on area estimation Gaussian Mixture Model is proposed. First, Gaussian mixture background model based on scale mapping is used for background detection and foreground clump of samples extraction. Then, the vehicle area estimation model is trained using the automatic samples collection and selection mechanism. After that, the model is used for foreground clumps selection, which can finally get the vehicle ROI. Experiments show that this method can provide more accurate vehicle ROI for the follow-up of vehicle detection, thus improves the real-time and reliability performance of vehicle detection.