Double-filter video multi-target tracking method based on IOU matching

• Main Authors: CHENG XIAOXUE, LIU JIANJUN, ZHANG YUAN, YANG JINLONG, MIAO JIANI
• Format: Patent
• Language: Chinese; English
• Published: 09.10.2020
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention discloses a double-filter video multi-target tracking method based on IOU matching, and belongs to the technical field of information processing. In multi-Bernoulli filtering, a correlation filter is introduced as a weak filter to sample particles, firstly, the convolution features of a VGG19 network with stronger generalization ability are extracted to train a plurality of correlation filters; a plurality of target state estimations are obtained by using the trained correlation filters, and finally, an obtained target state set is expanded to serve as a sampled particle set. Themethod can effectively reduce the label jump and track fragments, and improves the target tracking precision. 本发明公开了一种基于IOU匹配的双滤波器视频多目标跟踪方法,属于信息处理技术领域。本发明在多伯努利滤波中，引入相关滤波器作为弱滤波器来采样粒子，首先提取泛化能力较强的VGG19网络卷积特征训练多个相关滤波器，利用训练好的相关滤波器，得到多个目标状态估计，最后对得到的目标状态集进行扩充作为采样的粒子集。本发明能有效减少标签跳变和轨迹碎片，提高目标跟踪精度。