earticle

영어

Locating persons in crowded scenes is very difficult due to multi-resolution and complex environment. The other difficulty in pedestrian detection domain is the real time requirement, because the camera installed on the crossing road is in high definition. In this paper, we presented a multi-task pedestrian detection framework boosted by Bing feature. We firstly trained upright full-body, multi-person, half-body and head models, then we compute the object-ness score and generate 1000 proposals by Bing feature, and at last we apply different model to different aspect ratio of the detection proposals. The experiment results on the PASCAL VOC 2007 show that our method outperforms all the other methods and achieved lower miss rate than the state-of-the-art. The computation time cost is just the half of state-of-the-art method.