Traffic sign detection algorithm based on improved Tiny-YOLOv4

• Published in: Journal of physics. Conference series Vol. 2303; no. 1; pp. 12014 - 12019
• Main Authors: Zhang, Xiaoxue, Huang, Wei
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.07.2022
• Subjects: Accuracy; Algorithms; Feature extraction; Physics; Target detection; Traffic capacity; Traffic signs; Traffic speed
• ISSN: 1742-6588; 1742-6596; 1742-6596
• DOI: 10.1088/1742-6596/2303/1/012014

In order to balance the three requirements of high detection accuracy, fast detection speed and small model capacity, this paper proposes a traffic sign detection algorithm based on improved Tiny-YOLOv4. The backbone feature extraction network and feature pyramid network of original algorithm are improved, a new feature extraction module CRBottle is constructed and introduced into the network, and spatial pyramid pooling is added to enhance the ability of the network to extract features. Change to larger scale 52×52 and 104×104 prediction feature layers to improve the detection accuracy of small target traffic signs. The experimental results show that the improved algorithm achieves an average accuracy of 88.9% and a model memory capacity of 10.8MB. Compared with the original algorithm, the detection accuracy is increased by 13.4%, the model memory capacity is reduced by 11.8MB, the detection speed reaches 200FPS, and the comprehensive performance of the improved algorithm is better than other detection algorithms.