Enhanced pedestrian detection using deep learning based semantic image segmentation

摘要

Pedestrian detection and semantic segmentation are highly correlated tasks which can be jointly used for better performance. In this paper, we propose a pedestrian detection method making use of semantic labeling to improve pedestrian detection results. A deep learning based semantic segmentation method is used to pixel-wise label images into 11 common classes. Semantic segmentation results which encodes high-level image representation are used as additional feature channels to be integrated with the low-level HOG+LUV features. Some false positives, such as falsely detected pedestrians located on a tree, can be easier eliminated by making use of the semantic cues. Boosted forest is used for training the integrated feature channels in a cascaded manner for hard negatives mining. Experiments on the Caltech-USA pedestrian dataset show improvements on detection accuracy by using the additional semantic cues.