基于YOLOv3和ECO算法的电气化铁路接触网绝缘子的检测与跟踪

Detection and Tracking of Insulators of Electrified Railway Catenary Based on YOLOv3 and ECO Algorithm

在接触网绝缘子带电水冲洗装置视觉伺服控制系统的研究中,针对系统需连续获得绝缘子图像坐标的问题,提出一种针对绝缘子的目标检测跟踪一体化算法。该算法由YOLOv3目标检测器与多线程ECO-HC单目标跟踪器结合组成,其中,检测器用于绝缘子目标的初始检测和跟踪器学习,多线程跟踪器进行后续帧绝缘子目标的跟踪定位。为准确评估各跟踪器状态以决定是否调用检测器,引入了响应质量分数,并利用IOU等指标对检测结果进行综合分析。在绝缘子的检测跟踪试验中,算法基于单个CPU计算平台,其跟踪器的平均像素误差成功率为44.8%,平均重叠率成功率为49.7%,同时跟踪2个目标的平均速度为30.9 fps。结果表明,该算法能够在硬件资源有限的计算平台上准确、快速地完成绝缘子目标检测跟踪,满足控制系统要求。

In the research of the visual servo control system of the water washing device of the live catenary insulators,this paper,aiming at the problem that the system needs to continuously obtain the insulator image coordinates,proposes an integrated target detection and tracking algorithm for insulators.The algorithm is composed of YOLOv3 target detector and multi-threaded ECO-HC single target tracker,in which the detector is used for the initial detection of insulator targets and tracker learning,and the multi-threaded tracker performs tracking and positioning of the insulator targets in subsequent frames.In order to accurately evaluate the status of each tracker to determine whether to call the detector,this paper introduces the response quality score,and uses indicators such as IOU to comprehensively analyzes detection results.In the insulator detection and tracking experiment,the algorithm is based on a single CPU computing platform.The average pixel error success rate of the tracker is 44.8%,the average overlap rate success rate is 49.7%,and the average speed of tracking two targets at the same time is 30.9 fps.The results show that the algorithm can accurately and quickly complete the detection and tracking of insulator targets on a computing platform with limited hardware resources,and meet the requirements of the control system.