Analysis of vehicle and pedestrian recognition algorithm based on improved YOLO

Abstract

This paper expounds a vehicle pedestrian recognition algorithm based on improved YOLO. It analyzes the shortcomings of current algorithms and proposes improved schemes such as Feature Pyramid Network (FPN) and loss function adjustment. It extracts multi-layer image features through convolutional layers, performs feature fusion and multi-scale detection, and introduces Focal Loss instead of the original classification loss on the basis of the existing loss function to improve the current detection performance and solve the problem of inaccurate recognition of small targets.