Blind Lane Detection Algorithm Based on Improved YOLOv5s
Xueli Gong, Bingxu Pan, Zhongwen Yang, Wenyao Liu, Haoran Zheng · 2023
With the development of technology and society, there is a growing demand for Blindness aid devices. Detection algorithms for blind lanes are very important. In this paper, the network structure of YOLOv5s is used to greatly improve the recognition rate, and the CA is introduced to improve the network performance by considering both inter-channel relationships and long-distance location information without increasing the computational overhead. Changing PAN to BiFPN in the hope of achieving multi-scale feature fusion easily and quickly. In addition, considering that most of the Blindness aid devices are embedded, this paper also make lightweight improvements to the network replacing the traditional convolutional module with Ghost Module, which is intended to generate feature maps with less cost. Based on these improvements this paper improved the recognition rate of the network on the dataset to 98.3%, reduced model size by 26% and reduced FLOPS by 22% The FPS value reached 235Hz.