基于3D-MobileNetv2 的多目标实时跟踪框架

doi:10.3969/j.issn.1007-5461.2022.03.008

量子电子学报 ›› 2022, Vol. 39 ›› Issue (3): 364-372.doi: 10.3969/j.issn.1007-5461.2022.03.008

基于3D-MobileNetv2 的多目标实时跟踪框架

毛仁祥1, 常建华1;2*, 张树益1 李红旭1, 张露瑶1

( 1 南京信息工程大学江苏省大气环境与装备技术协同创新中心, 江苏南京210044; 2 南京信息工程大学江苏省气象探测与信息处理重点实验室, 江苏南京210044 )

收稿日期:2020-11-30 修回日期:2021-01-13 出版日期:2022-05-28 发布日期:2022-05-28
通讯作者: E-mail: jianhuachang@nuist.edu.cn E-mail: jianhuachang@nuist.edu.cn
作者简介:毛仁祥( 1996 - ), 江苏泰州人, 研究生, 主要从事点云处理, 目标检测与追踪等方面的研究。E-mail: 1203161293@qq.com
基金资助:
Supported by National Natural Science Foundation of China (国家自然科学基金, 61875089), Research and Practice Innovation Plan for Postgraduates in Jiangsu Province (江苏省研究生科研与实践创新计划, SJCX19-0308)

Multi-target real-time tracking system based on 3D-MobileNetv2

MAO Renxiang1, CHANG Jianhua1;2*, ZHANG Shuyi1, LI Hongxu1, ZHANG Luyao1

( 1 Collaborative Innovation Center of Atmospheric Environment and Equipment Technology, Nanjing University of Information Science & Technology, Nanjing 210044, China; 2 Jiangsu Key Laboratory of Meteorological Observation and Information Processing, Nanjing University of Information Science & Technology, Nanjing 210044, China )

Received:2020-11-30 Revised:2021-01-13 Published:2022-05-28 Online:2022-05-28

摘要/Abstract

摘要： 为了克服卡尔曼滤波在远距离以及遮挡情形下跟踪不准确的问题, 设计了一个多目标实时跟踪框架, 利用神经网络预测目标物体在三维空间里的状态, 使用匈牙利算法逐帧进行数据关联, 设计轨迹管理模块管理相应的轨迹, 实现多目标跟踪。相较于传统的框架, 该框架不需要在图像空间执行卡尔曼滤波, 不仅能在高帧速率准确跟踪遮挡目标, 而且对远距离目标也有优异的表现。以KITTI 数据集进行测试, 性能指标如下: 多目标跟踪准确度为79.22，多目标跟踪精确度为78.33, 多数丢失数为54.19, 多数跟踪数为13.21, ID 转变数为16, 运行速度为39 帧/秒。相较于传统的卡尔曼框架, 准确率提升了11%，并且抗遮挡性也有大幅度提升。

关键词: 图像处理, 多目标跟踪, 深度学习, 激光点云

Abstract: In order to overcome the inaccurate tracking problem of Kalman filter in the case of long distance and occlusion, a multi-target real-time tracking framework is designed. In the framework, neural network is used to predict the state of the target object in three-dimensional space, Hungarian algorithm is used for data association frame by frame, and the trajectory management module is designed to manage the corresponding trajectory to realize multi-target tracking. Compared with the traditional framework, this framework does not need to perform Kalman filtering in the image space. It can not only track occluded targets accurately at high frame rate, but also has excellent performance for long-distance targets. The performance of the framework on Kitti dataset is as follows: multi-target tracking accuracy is 79.22, multi-target tracking accuracy is 78.33, most of the lost number is 54.19, most of the tracking number is 13.21, ID transition number is 16, the running speed is 39 frames per second. Compared with the traditional Kalman framework, the accuracy of the designed framework is improved by 11%, and the anti occlusion performance is also greatly improved.

Key words: image processing, multi-target tracking, deep learning, laser point cloud

中图分类号:

TP391

毛仁祥, 常建华, 张树益李红旭, 张露瑶. 基于3D-MobileNetv2 的多目标实时跟踪框架[J]. 量子电子学报, 2022, 39(3): 364-372.

MAO Renxiang, CHANG Jianhua, ZHANG Shuyi, LI Hongxu, ZHANG Luyao. Multi-target real-time tracking system based on 3D-MobileNetv2[J]. Chinese Journal of Quantum Electronics, 2022, 39(3): 364-372.

参考文献

[1] Min Meng en, Zhou hang, Han Dan, et al. Overview of particle filter target tracking algorithm [J]. Computer engineering and application, 2019, 55 (05): 8-17 + 59
昝孟恩,周航,韩丹，等.粒子滤波目标跟踪算法综述[J].计算机工程与应用,2019,55(05):8-17+59.
[2] Meng Zhe, Yang Xu. Overview of target tracking algorithm [J]. Acta automatica Sinica, 2019,45 (07): 1244-1260
孟琭,杨旭.目标跟踪算法综述[J].自动化学报,2019,45(07):1244-1260.
[3] Li Xi, Zha Yufei, Zhang Tianzhu, et al. A survey of target tracking algorithms based on deep learning [J]. Acta Sinica Sinica, 2019, 24 (12): 2057-2080
李玺,查宇飞,张天柱，等.深度学习的目标跟踪算法综述[J].中国图象图形学报,2019, 24(12):2057-2080.
[4] Xiang Yi, Wang Yi, Zhang Jiachen, et al. Vehicle borne dual radar target position estimation method based on Lossless Kalman filter [J]. Optoelectronic engineering, 2019,46 (07): 102-110
向易,汪毅,张佳琛,等.基于无损卡尔曼滤波的车载双雷达目标位置估计方法[J].光电工程,2019,46(07):102-110.
[5] Liu Bin, Zheng Kaikai. Mean shift target tracking method based on four channel nonseparable wavelet [J]. Journal of quantum electronics, 2018,35 (01): 13-22
刘斌,郑凯凯.基于四通道不可分小波的均值漂移目标跟踪方法[J].量子电子学报,2018,35(01):13-22.
[6] Yang Xiaojun, pan Quan, Wang Rui, et al. Progress and Prospect of particle filter [J]. Control theory and application, 2006,23 (02): 261-267
杨小军,潘泉,王睿，等.粒子滤波进展与展望[J].控制理论与应用,2006,23(02):261-267.
[7] Cheng Yue, Li Jianzeng, Chu Lina, et al. Correlation filtering tracking algorithm based on model and scale updating [J]. Progress in laser and optoelectronics, 2018,55 (12): 297-303
成悦,李建增,褚丽娜,等.基于模型与尺度更新的相关滤波跟踪算法[J].激光与光电子学进展,2018,55(12):297-303.
[8] Ren S , He K , Girshick R , et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6):1137-1149.
[9] Lin T Y , Goyal P , Girshick R , et al. Focal Loss for Dense Object Detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, PP(99):2999-3007.
[10] Lee N , Weng X , Boddeti V N , et al. Visual Compiler: Synthesizing a Scene-Specific Pedestrian Detector and Pose Estimator[J]. 2016.
[11] Wu D L, Xue X H, Zhang D W, et al. Multi-Object Detection Method of 3D Point Cloud in Outdoor Scene Based on PointNet++[J]. Automation & Information Engineering, 2019.
[12] Shi S , Wang X , Li H . PointRCNN: 3D Object Proposal Generation and Detection from Point Cloud[J]. 2018.
[13] Zhou Y , Tuzel O . VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection[J]. 2017.
[14] Liu S X, Zheng J Y , Wang X , et al. Target Detection from 3D Point-Cloud using Gaussian Function and CNN[C]// 2019 34rd Youth Academic Annual Conference of Chinese Association of Automation (YAC). 2019.
[15] Ku J , Pon A D , Waslander S L . Monocular 3D Object Detection Leveraging Accurate Proposals and Shape Reconstruction[J]. 2019.
[16] Weng X , Kitani K . Monocular 3D Object Detection with Pseudo-LiDAR Point Cloud[J]. arXiv, 2019.
[17] Xiang Y , Alahi A , Savarese S . Learning to Track: Online Multi-object Tracking by Decision Making[C]// 2015 IEEE International Conference on Computer Vision (ICCV). IEEE, 2015.
[18] Wang S , Fowlkes C C . Learning Optimal Parameters for Multi-target Tracking with Contextual Interactions[J]. international journal of computer vision, 2016, 122:1-18.
[19] Frossard D , Urtasun R . [IEEE 2018 IEEE International Conference on Robotics and Automation (ICRA) - Brisbane, Australia (2018.5.21-2018.5.25)] 2018 IEEE International Conference on Robotics and Automation (ICRA) - End-to-end Learning of Multi-sensor 3D Tracking by Detection[J]. 2018:635-642.
[20] Simon M , Milz S , Amende K , et al. Complex-YOLO: Real-time 3D Object Detection on Point Clouds[J]. 2018.
[21] Weng X , Kitani K . A Baseline for 3D Multi-Object Tracking[J]. 2019.

基于3D-MobileNetv2 的多目标实时跟踪框架

Multi-target real-time tracking system based on 3D-MobileNetv2

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	於康杰, 方波, 李剑敏, 王震, 蔡晋辉, 邬佳璐, 何正龙 . 太赫兹成像三维块匹配自适应Canny边缘检测算法[J]. 量子电子学报, 2023, 40(4): 458-468.
[2]	葛宏义 , 王飞 , 蒋玉英 , 李丽 , 张元 , 贾柯柯 , . 基于宽度学习的太赫兹光谱图像小麦霉变识别研究[J]. 量子电子学报, 2023, 40(3): 360-368.
[3]	杜天宇, 高昆∗ , 吴朝. 光栅相衬 CT 中吸收信号环形伪影的去除方法[J]. 量子电子学报, 2023, 40(1): 40-47.
[4]	林惠祖刘伟涛孙帅杜隆坤常宸李月刚. 关联成像算法研究进展[J]. 量子电子学报, 2022, 39(6): 863-879.
[5]	林冰 , 樊学强 , 李德奎 , 彭志勇 , 郭忠义∗. 基于深度学习的散射光场成像研究进展[J]. 量子电子学报, 2022, 39(6): 880-898.
[6]	陈建明, , 曾祥津, , 钟丽云, , 邸江磊, ∗ , 秦玉文, ∗. 基于深度学习的图像配准方法研究进展综述[J]. 量子电子学报, 2022, 39(6): 899-926.
[7]	马慧敏 ∗ , 檀磊 , 张京会 , 张鹏飞 , 宁孝梅 , 刘海秋 , 高彦伟 . 基于深度学习的合成孔径成像系统共相误差检测研究综述[J]. 量子电子学报, 2022, 39(6): 927-941.
[8]	王孝艳, 王志远, 陈子阳, 蒲继雄∗. 基于深度学习技术从散斑场中识别多涡旋结构的轨道角动量[J]. 量子电子学报, 2022, 39(6): 955-961.
[9]	吴琼, 马雷∗. 一种基于 LoG 算子的量子图像边缘检测算法[J]. 量子电子学报, 2022, 39(5): 720-727.
[10]	李天秀, 石磊∗, 王俊辉, 李佳豪. 基于深度学习的量子信号大气衰减系数预测[J]. 量子电子学报, 2022, 39(5): 786-794.
[11]	杨一杉, 梁华为, 姚瑶, 陈帅∗. 金属遮挡条件下光子层析快速重构方法研究[J]. 量子电子学报, 2022, 39(4): 558-565.
[12]	张杨∗, 程正东, 朱斌. 基于改进 Faster R-CNN 的无人机小目标检测算法[J]. 量子电子学报, 2022, 39(3): 354-363.
[13]	臧一鸣, 朱尚超, 魏战红∗, 刘志飞, 林小竹, 孙文韬. 一种量子图像伪彩色编码方法[J]. 量子电子学报, 2022, 39(3): 343-353.
[14]	洪富祥, 陈冲, 丘仲锋∗. 一种基于机器视觉的无人机同心圆靶精准降落方法[J]. 量子电子学报, 2021, 38(3): 307-315.
[15]	陈儒, 杨义勤, 秦凡凯, 孟昭晖, 苗昕扬, 赵昆, 詹洪磊∗. 样本内部缺陷的太赫兹光谱成像与算法优化[J]. 量子电子学报, 2021, 38(3): 265-271.