Pedestrian detection framework based on magnetic regional regression
Yao Li, Bofan Wang · IET Image Processing · 2019
In recent years, pedestrian detection has become one of the core issues in self‐driving car and automatic alert in video surveillance. However, the efficiency and accuracy are not ideal in multi‐pedestrian scenarios because of the occlusion by other pedestrian or non‐human objects. Here, the authors design a pedestrian‐detection framework based on region proposal network and convolutional neural network. Magnetic region regression strategy was proposed to reduce locating errors and false detection in the process of region regression. Meanwhile, semantic segmentation is integrated into authors’ framework to improve the accuracy of classification. Additionally, the authors used soft‐ non‐maximum suppression (NMS) to reduce the impact of NMS threshold on the model. Authors’ framework has average miss rate of 6.94% on Improved Caltech‐USA dataset. The experiments show that authors’ framework achieves significant advantages.