一种遗留物品检测算法在分布式模型中的应用

摘要

Dropped object detection is an important research topic in the fields of artificial intelligence and pattern recognition.In this paper,an algorithm based on regional information is proposed to detect dropped objects in the video.The algorithm consists of three steps:constructing and updating dual-direction background models to obtain and maintain stationary objects,tracking each moving pedestrian by mean-shift and using pixel-based regional information to detect dropped objects.Meanwhile our algorithm is realized in a distributed real-time video detection model.The ex-perimental results demonstrate that our algorithm can detect dropped objects quickly and accurately on various experi-mental video scenes.And the proposed distributed model is more appropriate for large-scale real-time video automation detection than the centralized model.%遗留物品的视频检测是人工智能和模式识别领域的一个重要的研究课题,本文提出了一种基于区域信息的算法来检测视频中的遗留物品.该算法由三个步骤所组成:通过构建并更新一个双重方向的背景模型来获取和维持视频中的静止物体;使用 mean-shift 算法跟踪视频中每一个移动的行人;使用基于像素级别的区域信息对遗留物品进行检测.同时将该算法在一个分布式的实时视频检测模型中得以实现.实验结果表明,该算法在多种不同的实验视频场景中可以快速且精确地检测出视频中出现的遗留物品.并且本文提出的分布式模型与集中式模型相比,更适用于大规模的实时视频自动化监控和检测.