项目名称: 移动终端视频目标快速识别技术研究
项目编号: No.61461039
项目类型: 地区科学基金项目
立项/批准年度: 2015
项目学科: 无线电电子学、电信技术
项目作者: 马颖东
作者单位: 内蒙古大学
项目金额: 43万元
中文摘要: 基于机器学习的视频物体识别算法在近些年得到了快速发展,但由于算法的计算复杂度较高,目前还没有在移动终端上实现视频物体快速识别的成功应用。我们立足于提高特征提取和分类器训练的效率来解决这个问题。本项目的研究重点是:(1)设计实现描述视频对象的新特征,提高特征计算的效率。(2)将局部特征和区域特征融合,在不同尺寸区域以及不同长方比区域中计算特征值,使组合特征可以更好地描述不同类型目标在较大区域内的特征分布情况。(3)引入缩放因子,简化相邻缩放尺度图像中的特征提取,找到算法效率和算法准确率之间的最佳平衡点,在保证识别准确率的同时提高系统的识别效率。本项目研究的关键技术拥有广泛的应用前景。通过与企业合作可以在智能安防、智能交通、电子商务等领域实现巨大的市场价值。
中文关键词: 计算机视觉;特征提取;目标检测
英文摘要: Video object recognition based on machine learning technologies has been developed rapidly in these years. However, due to the computational complexity of recognition algorithms, there is no successful video object recognition application for mobile devices so far. We try to solve this problem by improving efficiency of feature extraction and classifier training. In this project, research works focus on three topics. Firstly, design and implement new features to accelerate feature extraction. Secondly, integrate local features and region features so that these multi-level features cover larger area of target objects. Finally, we use scale factor to simplify feature calculation between two adjacent scale images. The key problem is to find an optimal tradeoff between accuracy and efficiency. The key technologies of this project have extensive application prospect and,they can be implement to realize huge market values by means of cooperation with business partners.
英文关键词: Computer Vision;Feature Extraction;Object Detection