Rapid Pedestrian Detection Based On Movement Trend
Ruohua Li, Taihong Wang · 2016
This paper presents a movement trends based approach for pedestrian detection aiming at reducing the consumption of feature calculation caused by sliding windows.A new approach to predict the location of pedestrian is proposed by combining the movement trend of objects, extracted by improved background segmentation algorithm, with Kalman filter.The keypoint descriptor BRISK (Binary Robust Invariant Scalable Keypoints) is presented to verify the predicted location and make it reliable.Experiment results on PETS dataset report that the algorithm is 10.9 times faster than SVM+HOG method and keep a better accuracy at the same time.