基于视觉的交通场景目标检测与跟踪技术研究
Research on Vision-Based Traffic Scene Object Detection and Tracking Technology
张涌 1周爱博 1黄林雄 1赵奉奎1
作者信息
- 1. 南京林业大学汽车与交通工程学院,江苏 南京 210037
- 折叠
摘要
针对通用目标检测算法在交通场景下检测多目标时存在的小目标漏检和误检问题,提出了一种适用于交通场景下的目标检测与跟踪算法.基于K-means++聚类算法计算出适合交通场景下目标检测的锚定框(Anchor Box)和深度可分离卷积的思想优化目标检测算法.其次,利用匈牙利匹配将卡尔曼滤波的预测值和目标检测的实际值相关联匹配,以实现目标跟踪.实验结果证明,改进后的目标检测与跟踪算法较原始算法在mAP 方面提升了 2.3%,达到了 99.13%;速度提升了20%,达到了 30 帧/秒,对交通场景下的特定目标均实现了准确检测跟踪,保持了较好的实时性.
Abstract
A target detection and tracking algorithm suitable for traffic scenes is proposed to address the issues of small target missed detection and false detection in detecting multiple targets using general object detection algorithms.First,an anchor box suitable for object detection in traffic scenes was calculated based on the K-means++ clustering algorithm,and depthwise separable convolution was used to optimize the object detection algorithm.Second-ly,the Hungarian matching algorithm was used to realize the correlation matching between the predicted valueofthe Kalman filter algorithm and the actual value,so as to realize the target tracking.The experimental results show that compared with the original algorithm,the mAP of the improved object detection and tracking algorithm in this paper is improved by 2.3%to 99.13%,and speed is increased by 20%to 30 frames per second.The targets in the traffic scene are accurately detectedandtracked,maintaining the great real-time performance.
关键词
目标检测/目标跟踪/卡尔曼滤波/匈牙利匹配Key words
Object detection/Target tracking/Kalman filter/Hungarian match引用本文复制引用
基金项目
江苏省重点研发计划(产业前瞻与共性关键技术)(BE2017008-2)
出版年
2024
NETL
NSTL
万方数据