基于YOLOv3和Deep SORT的草原牛跟踪系统

Grassland cattle tracking system based on YOLOv3 and Deep SORT

设计了一种基于深度学习算法的草原牛跟踪系统。融合YOLOv3目标检测算法与Deep SORT目标跟踪算法实现对草原牛的检测跟踪,结合比例—积分—微分(PID)算法控制云台(PTZ)摄像头稳定跟随草原牛转动。在内蒙古苏尼特左旗牧场进行现场实验测试,实验结果表明:系统运行稳定,对草原牛检测准确率较高,跟踪效果较好,可以实现未检测到草原牛时自动巡航、对多只草原牛自动跟踪、以及指定跟踪单只草原牛的功能。

A grassland cattle tracking system based on deep learning algorithms is designed.The system fuses YOLOv3 target detection algorithm and Deep SORT target tracking algorithm to detect and track grassland cattle, and combines the proportional-integral-differential(PID)algorithm to control the PTZ camera to follow the grassland cattle stably.The system performs field experiments on the Sunit Zuoqi pasture in Inner Mongolia.The experimental results show that the system runs stably with high accuracy in detecting steppe cattle, and has a good tracking effect.It can achieve automatic cruise when no steppe cattle are detected, and multiple steppe automatic cattle tracking, and the function to track a specified single steppe cattle.