期刊文献+

基于跨尺度特征聚合网络的行人检测与跟踪 被引量:2

Pedestrians detection and tracking based on Trans-scale Feature Aggregation Network
下载PDF
导出
摘要 针对行人检测跟踪过程中目标空间尺度差异大、小目标漏检率高,以及跟踪稳定性差的问题,提出一种基于跨尺度特征聚合网络(TFACN)的行人多目标检测跟踪方法。该方法选取CenterNet网络作为目标检测器,并设计跨尺度特征聚合模块(TFA),对高、低层特征信息进行有效融合,实现对卷积层特征的增强表示,有效地降低虚警率、小目标的漏检率和计算复杂度;其次在高精度识别的基础上,结合自适应卡尔曼滤波,通过实时计算残差信息在线更新观测噪声协方差,对行人运动状态进行自适应估计,提高行人跟踪的位置估计精度。试验结果表明,本文算法相较于现有的算法,可以在较低的计算成本下获得更高的精度。 Aiming at the problems of large spatial scale difference of targets,high missing rate of small target pedestrians and poor tracking stability in pedestrians detection and tracking process,a pedestrian multi-target detection and tracking method based on Trans-scale Feature Aggregation CenterNet(TFACN)is proposed.Firstly,CenterNet network is selected as the target detector,and the Trans-scale Feature Aggregation module(TFA)is designed to effectively fuse the high and low level feature information,realize the enhanced representation of the feature of the convolutional layer,and effectively reduce the false alarm rate,the missing rate of small targets and the computational complexity.Secondly,on the basis of high precision identification,combined with adaptive Kalman filter,the pedestrians state is estimated adaptively through real-time residual information computation to update observation noise covariance,and the accuracy of pedestrians tracking position estimation is improved.Experimental results show that compared with the existing algorithms,the proposed detection and tracking algorithm can achieve higher accuracy at lower computational cost when tested on Caltech pedestrian dataset.
作者 刘康安 肖永超 LIU Kang'an;XIAO Yongchao(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处 《智能计算机与应用》 2022年第8期27-32,42,共7页 Intelligent Computer and Applications
关键词 行人检测 特征聚合 多目标跟踪 自适应滤波 pedestrians detection feature aggregation mutiple object tracking adaptive filtering
  • 引文网络
  • 相关文献

参考文献2

二级参考文献16

  • 1杨小军,潘泉,张洪才.基于粒子滤波和似然比的联合检测与跟踪[J].控制与决策,2005,20(7):837-840. 被引量:14
  • 2彭宁嵩,杨杰,刘志,张风超.Mean-Shift跟踪算法中核函数窗宽的自动选取[J].软件学报,2005,16(9):1542-1550. 被引量:165
  • 3宋慧波,高梅国,田黎育,毛二可,顾文彬.一种基于动态规划法的雷达微弱多目标检测方法[J].电子学报,2006,34(12):2142-2145. 被引量:16
  • 4刘建成,王雪松,刘忠,肖顺平,王国玉.基于Wigner-Hough变换的LFM信号检测性能分析[J].电子学报,2007,35(6):1212-1216. 被引量:34
  • 5Salmond D J, Birch H. A particle filter for track-beforedetect[C]. IEEE Proc of the American Control Conf. Washington, 2001: 3755-3760.
  • 6Ristic R, Arulampalam S, Gordon N. Beyong the Kalman filter: Particle filters fo tracking application[M]. Boston- London: Artech House, 2004.
  • 7Boers Y, Driessen J N. Multitarget particle filter track before detect application[J]. IEE Proceedings-Radar, Sonar and Navigation, 2004, 151(6): 351-357.
  • 8Rutten M G, Gordon N, Maskell S. Recursive track before detect with target amplitude fluctuations[J]. IEE Proceedings-Radar, Sonar and Navigation, 2005, 152(5): 345-352.
  • 9Su H T, Wu T P, Liu H W, et al. Rao-Blackwellised particle filter based track before detect algorithm[J]. IET Signal Processing, 2008, 2(2): 169-176.
  • 10Arulampalam M S, Maskell S, Gordon N, et al. A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking[J]. IEEE Trans on Signal Processing, 2002, 50(2): 174-188.

共引文献10

同被引文献15

引证文献2

;
使用帮助 返回顶部