基于连续离散问题联合求解和群组分析的多目标跟踪技术研究

Multi-Object Tracking Based on Continuous-Discrete Problem Solving and Group Analysis

摘要: 多目标跟踪技术通过对不同目标之间的相互社会关系进行建模，改善单个目标的跟踪性能，并且快速检测和预判场景中可能发生的群体类突发事件。现有的多目标跟踪技术虽在数据关联和轨迹估计上取得平衡，但依然存在诸多问题。本文介绍通过背景建模提取出的场景信息分析并识别目标来约束多目标跟踪，将数据关联和轨迹估计这两个连续和离散的经典子问题结合到统一的框架中求解；与此同时，还提出了基于群组聚类的行为建模策略，得到的语义信息提供相邻目标和轨迹之间的约束，有助于改善跟踪结果。实验表明，本文提出的策略相比经典的多目标跟踪算法准确性更高。

Abstract: Multi-object tracking technique can improve the tracking performance of a single target by mod-eling the mutual social relationship between different targets, and emergencies groups were quickly detected and predicted scenarios that may occur. Although the existing multi-object tracking technology can balance the data association and trajectory estimation, there are still many problems. This paper introduces the scene information extracted from the background modeling and object recognition to constrained multi-target tracking, and combined into a unified framework for solving with the data association and track estimation of both discrete and continuous classic sub-problems; at the same time, we propose the behavior modeling strategy based on group clustering, whose semantic information places the restriction between adjacent target and its track, which can help improve the trace results. The results of the experiment show that the proposed method is more accurate than the classical multi-target tracking algorithm.