Blind obstacle avoidance system based on improved YOLOv3
Fang Li, Junwu Xu, Yang Li, Changde Li · 2021 IEEE 2nd International Conference on Information Technology, Big Data and Artificial Intelligence (ICIBA) · 2021
In order to facilitate outdoor travel for blind people, this paper proposes a blind person's obstacle avoidance system based on the improved YOLOv3 algorithm. The system design is based on C/S architecture, the user device consists of STM32F103, serial module, camera module, 4G module, etc.; the server software system mainly includes two parts: data sending and receiving and target detection. First of all, the pictures taken by the camera module are stored in the data buffer, and the 4G module is used to transmit the picture data to the server, after the server receives the picture data, it will be reduced to pictures, and then the trained F-YOLOv3 model is used to detect the pictures, and if an obstacle is detected, the result will be returned to STM32F103, and STM32F103 will output this result to the user by voice, so as to achieve the purpose of obstacle avoidance. The STM32F103 will output this result to the user, thus achieving the purpose of obstacle avoidance. In this paper, we propose an improved F-YOLOv3 based on YOLOv3, which increases the original 3 feature scales in YOLOv3 to 4, thus reducing the miss detection rate of small target objects; replaces some convolutional layers in Darknet53 with improved deep separable convolutional blocks. The depth-separable convolutional block in Darknet53 is replaced with an improved depth-separable convolutional block, which speeds up the detection speed.