Multi-Scale Pedestrian Detection through Feature Fusion in Faster RCNN with VGG16

Abstract

Addressing the issue of multi-scale challenges in pedestrian detection, this study introduces a novel pedestrian detection model that integrates Faster R-CNN with multi-scale feature fusion. The model employs VGG16 as the foundational feature extraction network, which is utilized for feature fusion. Subsequently, the model generates features characterized by high resolution and rich semantics, which are then used for prediction, followed by classification and regression processes. Experimental results on a custom pedestrian dataset indicate that the proposed method enhances the capability of multi-scale pedestrian detection to a significant degree.