Real-time pedestrian detection via hierarchical convolutional feature

With the development of pedestrian detection technologies, existing methods cannot simultaneously satisfy high quality detection and fast calculation for practical applica-tions. Therefore, the goal of our research is to balance of pedestrian detection in aspectsof the accuracy and efficiency, then get a relatively better method compared with currentadvanced pedestrian detection algorithms. Inspired from recent outstanding multi-categoryobjects detector SSD (Single Shot MultiBox Detector), we proposed a hierarchical convo-lution based pedestrians detection algorithm, which can provide competitive accuracy ofpedestrian detection at real-time speed. In this work, we proposed a fully convolutional net-work where the features from lower layers are responsible for small-scale pedestrians andthe higher layers are for large-scale, which will further improve the recall rate of pedestrianswith different scales, especially for small-scale. Meanwhile, a novel prediction box with asingle specific aspect ratio is designed to reduce the miss rate and accelerate the speed ofpedestrian detection. Then, the original loss function of SSD is also optimized by eliminat-ing interference of the classifier to more adapt pedestrian detection while also reduce thetime complexity. Experimental results on Caltech Benchmark demonstrates that our pro-posed deep model can reach 11.88% average miss rate with the real-time level speed of 20fps in pedestrian detection compared with current state-of-the-art methods, which can bethe most suitable model for practical pedestrian detection application