基于YOLO-V4的矿用巡检机器人无标定视觉伺服控制系统

Non-calibrated Visual Server Control System for Mining Patrol Robot Based on YOLO-V4

针对传统的矿用巡检机器人视觉伺服多是在标定条件下实现,标定参数精度较差,无标定视觉伺服控制系统因其具有更好的适应性、灵活性,越来越得到广泛的认可。基于YOLO-V4算法,采用CSPDarknet53网络结构,对矿用巡检机器人无标定视觉伺服控制系统进行设计实验,并与YOLO-V3,SSD,AlexNet这3种模型进行分析比较。结果表明:基于YOLO-V4的矿用巡检机器人无标定伺服控制技术切实可行,其识别精度较高,实时性良好,满足当前矿业监测的需求,能够进一步提升作业的安全性。

For the traditional mine inspection robot,the visual servo system is mostly implemented under calibration conditions,and the calibration parameters have poor accuracy.The non-calibration visual servo control system is more and more widely recognized because of its better adaptability and flexibility.Based on YOLO-V4 algorithm and using CSPDarknet53 network structure,designs and tests a non-calibrated visual servo control system for mining patrol robot,and compares it with three models:YOLO-V3,SSD and AlexNet.The results show that the non-calibration server control technology of the mining patrol robot based on YOLO-V4 is feasible,its recognition accuracy is high,its real-time performance is good,it meets the current requirements of mining monitoring,and it can further enhance the safety of operation.