基于推理系统的行人跟踪优化算法  

Optimization Algorithm for Pedestrian Tracking Based on Inference System

作　　者：汤义 黄建军[2] 易娇君 TANG Yi;HUANG Jian-jun;YI Jiao-jun(Guangzhou College of Commerce,Guangzhou Guangdong 511363,China;Guangdong Key Laboratory of Intelligence Information Processing,Shenzhen University,Shenzhen Guangdong 518060,China)

机构地区：[1]广州商学院,广东广州511363 [2]深圳大学广东省智能信息处理重点实验室,广东深圳518060

出　　处：《计算机仿真》2025年第2期221-225,446,共6页Computer Simulation

基　　金：国家自然科学基金面上项目(62076165);广东省教育科学规划课题(2022GXJK367)。

摘　　要：针对如何使用已有的视觉分析结果完成更为复杂的检测和跟踪,改善复杂场景中行人跟踪的性能和鲁棒性等相关问题,提出了一种基于MPEG-7标准描述的行人跟踪算法。首先,根据MPEG-7多媒体描述标准提出并实现一种行人描述模型,对已有的行人检测跟踪结果进行描述。在描述模型的基础之上,提出了一种基于推理系统的行人检测与跟踪算法。算法的核心包含三个部分,推理规则,行为规则和命题规则。通过命题规则的判断,推理规则的推理和行为规则的操作,上述方法克服了目标跟踪不连续以及复杂情况下的聚集分离等问题。同时,跟踪的结果反馈给描述信息,通过描述信息和跟踪结果的相互作用,提高了跟踪的效果,最终实现了复杂环境下的行人跟踪。最后的实验结果说明了所提算法正确性。This article proposes a pedestrian tracking algorithm based on the MPEG-7 standard to address issues related to how to use existing visual analysis results to achieve more complex detection and tracking,and improve the performance and robustness of pedestrian tracking in complex scenes.Firstly,a pedestrian description model is proposed and implemented based on the MPEG-7 multimedia description standard,which describes the existing pedestrian detection and tracking results.On the basis of describing the model,a pedestrian detection and tracking algorithm based on inference system is proposed.The core of this algorithm consists of three parts:inference rules,behavior rules,and proposition rules.By judging propositional rules,reasoning rules,and operating behavioral rules,this method overcomes problems such as discontinuous target tracking and aggregation separation in complex situations.At the same time,the tracking results are fed back to descriptive information,and the interaction between descriptive information and tracking results improves the tracking effect,ultimately achieving pedestrian tracking in complex environments.The final experimental results demonstrate the correctness of the algorithm proposed in this paper.

关 键 词：推理系统 行人跟踪 深度学习 描述模型 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]