Pedestrian Detection and Segmentation Method Based on Mask R-CNN

Abstract

Pedestrian detection and segmentation are difficult to perform well under the same model, because of the variability in the complex scene. In this paper, a multi-scale feature extraction convolution neural network based on Mask R-CNN is proposed. In the feature pyramid, the features of shallow, medium and deep are mixed together to get feature maps at different scales, which provides us with comprehensive feature information. And we set a lot specific anchors to get a full range of proposed areas in FPN to lift detection accuracy rates and make high-precision segmentation. Finally, the ROI Align extracts features from each proposal to get the detection and segmentation result. Our method test on the INRIA dataset, which shows that the pedestrian detection and segmentation algorithm based on Mask R-CNN performs well and performs better than other algorithm.