Traffic Sign Detection and Recognition Using Dense Connections in YOLOv4

Abstract

A self-driving car is a growing technology in India where detection of traffic sign in an unconstrained environment is a challenging task due to its small size. With the development of deep neural networks, many models for object detection have been developed. In this work, we have used a single-stage detection model YoloV4 with further improvements in detection neck for detection of traffic signs. We have used dense connections in place of normal connections of the model for better feature propagation. This improves the accuracy with less inference time. We have conducted our experiments on bench-marked Chinese traffic sign dataset, Tshigua-Tenscent 100K dataset (TT-100K). We have achieved accuracy of 94.30% with 32 FPS. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.