基于区分性模型学习的综合在线多物体检测、跟踪和分割

项目名称： 基于区分性模型学习的综合在线多物体检测、跟踪和分割

项目编号： No.61303178

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 兴军亮

作者单位： 中国科学院自动化研究所

项目金额： 25万元

中文摘要： 作为计算机视觉领域的热点研究问题，视频中物体的检测、跟踪和分割技术是绝大多数智能监控系统的前提和基础，具有非常重要的研究价值和广阔的应用前景。本课题提出通过设计一些新型的区分性模型及其学习算法，实现对视频中的多个物体同时进行在线检测、跟踪和分割。研究内容包括：（1）基于区分性模型学习的在线多物体检测算法；（2）基于区分性模型学习的在线多物体跟踪技术；（3）基于区分性模型的学习的在线多物体分割技术；（4）基于多物体检测、跟踪和分割算法的视觉监控原型系统设计。本课题的研究目标是通过一系列高效和鲁棒的多物体检测、跟踪和分割算法，来提高现有视觉监控系统的有效性和实用性，从而实现一个智能化视觉监控原型系统，并将其用于公共场所的人群和车辆监控。该项研究不仅能够促进视频中物体检测、跟踪和分割等相关技术的发展，而且对于构建平安城市和维护社会稳定具有非常重要的现实意义。

中文关键词： 物体检测；物体跟踪；物体分割；区分性模型；视觉监控

英文摘要： Among the most active research fields in computer vision, object detection, tracking and segmentation in videos are of fundamental importance to most intelligent visual surveillance system, and has significant academic research value and broad application potentials. This project proposes to design some novel discriminative models as well as their learning methods, and develop a system which can online simultaneously detect, tract and segment multiple objects in videos. The project will focus on four key points: (1) discriminative model based online multiple object detection; (2) discriminative model based online multiple object tracking; (3) discriminative model based online multiple object segmentation; (4) visual surveillance prototype system designation based on multiple object detection, tracking and segmentation. The research subject of this project is, through several efficient and robust multiple object detection, tracking and segmentation algorithms, to implement a prototype visual surveillance system applied on crowd and vehicle monitoring, which can improve the effectiveness and practicality of existing systems. This project will not only improve the development of video object detection, tracking and segmentation techniques, but also has great practical significances for building safe cities and main

英文关键词： object detection；object tracking；object segmentation；discriminative model；visual surveillance