基于压缩感知的鲁棒视频运动检测和跟踪技术的研究

项目名称： 基于压缩感知的鲁棒视频运动检测和跟踪技术的研究

项目编号： No.61262067

项目类型： 地区科学基金项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 高赟

作者单位： 云南大学

项目金额： 47万元

中文摘要： 视频运动检测和跟踪是计算机视觉领域的研究热点之一，但是实际复杂场景下鲁棒运动检测和跟踪仍存在以下主要挑战：1）需要对像素分布进行假设，但并不总是符合实际场景；2）需要对光照突变、阴影特别处理或后处理，缺乏适应性算法； 3)跟踪过程中背景干扰、表观变化、局部遮挡情况影响跟踪效率。压缩感知是一种新兴信号采集理论，无需对像素分布进行任何假设，可以直接对运动检测和跟踪关注的稀疏性前景信号进行采样，并重构出前景目标。本项目在建立基于压缩感知提取稀疏性前景信号模型的基础上,通过构造抑制光照突变和阴影的测量矩阵以得到鲁棒运动检测算法，进而通过有效特征提取和目标稀疏表述来实现抑制背景干扰、表观变化、局部遮挡的鲁棒目标跟踪，其研究将促进压缩感知理论与视频运动跟踪技术的融合,并为视频运动检测跟踪中挑战性问题提供新的解决思路。预期在国内外重要期刊及会议上发表论文12篇，并申请1-2项专利或软件著作权登记。

中文关键词： 运动检测；目标跟踪；压缩感知；测量矩阵；

英文摘要： Video-based motion detection and tracking is one of research hotspots in the field of Computer Vision. There still exist several main challenges about robust motion detection and tracking, due to the complexities of real application scene. These challenges are that supposing about pixel distribution in traditional algorithms do not always accord with real scenes, that traditional algorithms lack the adaptive method to sudden illumination change and shadow, and that tracking efficiency is affected by background interfere, appearance variation and partial occlusion. Compressive sensing is a new type of sampling theory, which need not suppose pixel distribution, and can directly sample foreground signals and reconstruct foreground motion objects. This proposal intends to build a model about sampling sparse foreground signals based on compressive sensing, gain robust algorithms of motion detection by constructing an effective measurement matrix, restraining sudden illumination change and shadow, and implement robust tracking by effectively extracting features and representing object features based sparse respresentation, restraining background interfere, appearance variation and partial occlusion. Related researching work can promote the fusion of compressive sensing theory and motion detection, and provide some new

英文关键词： Motion Detection；Object Tracking；Compressive Sensing；Measurement Matrix；