交通运输系统工程与信息

交通运输系统工程与信息 ›› 2022, Vol. 22 ›› Issue (6): 85-94.DOI: 10.16097/j.cnki.1009-6744.2022.06.009

• 智能交通系统与信息技术 • 上一篇    下一篇

一种基于物影匹配算法的道路小目标跟踪方法

项新建*1，胡海斌1，姚佳娜2，丁祎1，郑永平1，金立3   

  1. 1. 浙江科技学院，自动化与电气工程学院，杭州 310023；2. 浙江省交通运输科学研究院，杭州 310023； 3. 浙江小桥流水环境科技有限公司，杭州 310023
  • 收稿日期:2022-08-06 修回日期:2022-09-29 接受日期:2022-10-10 出版日期:2022-12-25 发布日期:2022-12-22
  • 作者简介:项新建(1964- )，男，浙江永康人，教授。
  • 基金资助:
    浙江省自然科学基金(LY19F030004)；浙江省基础公益研究计划项目(LGF22F030005)；浙江省重点研发计划项目(202206)。

A Road Small Target Tracking Approach Based on Object-shadow Matching Algorithm

XIANG Xin-jian*1, HU Hai-bin1, YAO Jia-na2, DING Yi1, ZHENG Yong-ping1, JIN Li3   

  1. 1. School of Automation and Electrical Engineering, Zhejiang University of Science and Technology, Hangzhou 310023, China; 2. Zhejiang Scientific Research Institute of Transport, Hangzhou 310023, China; 3. Zhejiang Bridge Water Sci-tech of Environment Co. Ltd, Hangzhou 310023, China
  • Received:2022-08-06 Revised:2022-09-29 Accepted:2022-10-10 Online:2022-12-25 Published:2022-12-22
  • Supported by:
    Natural Science Foundation of Zhejiang Province, China；Public Welfare Technology Application Research Project of Zhejiang Province；Key R & D Program Project of Zhejiang Province, China

PDF

摘要： 针对复杂道路环境下小目标抛洒物难以被检测与跟踪，且落地静止像素易被纳入背景而导致目标丢失等问题，本文提出一种基于物影匹配算法(OSMA)的道路小目标跟踪方法来解决上述问题。首先，采用混合高斯模型对背景进行建模以此获取前景图像，对前景图像采用膨胀、腐蚀等形态学处理；其次，根据道路抛洒物移动时在前景产生独立的物与影双轮廓特征，进行连续帧前景轮廓的物影匹配，并将连续的匹配结果定为疑似抛洒物；最后，采用多帧质心偏移法判断疑似抛洒物是否处于静止状态，并对运动状态中的疑似抛洒物进行位置判定与帧间轮廓匹配，从而实现抛洒物的逐帧跟踪。基于大量实验表明：本文所提出的OSMA与核化相关滤波器(KCF)、辨别尺度空间跟踪器(DSST)、背景感知相关滤波器(BACF)、深度注意力跟踪器(DAT)、视觉跟踪时空变换器(SATRK)等跟踪器相比，准确性更优越，可较好解决道路复杂场景下各类抛洒物的跟踪问题；复杂背景、快速旋转的小目标场景中跟踪表现优异，具有良好的跟踪尺度，实时性满足预期要求。

关键词: 智能交通, 道路抛洒物跟踪, 物影匹配, 小目标抛洒物, 复杂道路环境, 滤波跟踪

Abstract: In complex road environments, small target throwing objects are difficult to be detected and tracked, and the falling stationary pixels are easily included in the background and thus it can lead to target loss. In this paper, a road small target tracking approach based on an object-shadow matching algorithm (OSMA) is proposed. A hybrid Gaussian model is used to model the background to obtain the foreground image, and morphological processing such as expansion and erosion is applied to the foreground image. According to the dual contour feature of the independent object and shadow in the foreground when the throwing object moves, object-shadow matching of the foreground contour is performed in consecutive frames, and the consecutive matching results are designated as the suspected throwing object. Finally, the multiframe center-of-mass offset method is used to determine whether the suspected throwing object is in the stationary state, and to determine the position of the suspected throwing object in the motion state and match the contour between frames to achieve the frame-by-frame tracking of the throwing object. Based on a large number of experiments, the proposed OSMA has better accuracy than the KCF, DSST, BACF, DAT, SATRK, and other trackers. It can better solve the tracking problem of various types of spills in complex road scenes, and has excellent tracking performance in small target scenes with complex backgrounds and fast rotation, with good tracking scale and real-time performance to meet the expected requirements.

Key words: intelligent transportation, road abandoned object tracking, object-shadow matching, small target throwing objects, complex road environment, filter tracking

中图分类号: