Pedestrian detection based on multi-scale feature fusion
Lincai Huang, Zhiwen Wang, Xiaobiao Fu · 2022 IEEE 5th Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC) · 2022
To solve the problem of a large difference in target size in pedestrian detection, which leads to high pedestrian false detection rate and high miss detection rate of small-scale pedestrians, a multi-scale feature fusion method based on RetinaNet is proposed. After feature enhancement by extracting features from the backbone network, three branches of different scales are formed, which are effectively fused with the corresponding feature layer to further enrich the target information, and then detect pedestrians of different scales. Test on the open dataset shows that compared with the original RetinaNet algorithm, it can detect more pedestrians, especially small-scale pedestrians, and the model detection performance is better.