Locally Assembled Binary feature with feed-forward cascade for pedestrian detection in intelligent vehicles

Detecting pedestrians in images is a challenging task, especially for the intelligent vehicle environment where there is a real-time requirement which limits the computational complexity of algorithms. In this paper, we demonstrate a near real-time and robust pedestrian detection system in the context of intelligent vehicle. This is achieved by integrating Locally Assembled Binary (LAB) features with a feed-forward cascade structure. LAB feature, which comprises several neighboring binary Haar features with a similar idea to Local Binary Pattern (LBP), is not only efficient in evaluation, but also very discriminative for pedestrian/non-pedestrian classification. Furthermore, a feed-forward cascade structure, which can exploit both the stage-wise and the cross-stage information, is presented to build on an efficient detector. Experimental results demonstrate the effectivity and efficiency of the proposed method.