摘要
针对运动目标检测不适合实时性应用场合的问题,提出了一种将无监督特征学习和显著性检测相结合的地面车辆目标检测算法。通过学习得到表示车辆目标的局部特征并进行编码,根据这些特征对整个图像进行显著性检测,获得候选目标区域。通过相关分析去除那些高度相关的特征,有效抑制背景,突出显著对象。
Aimed at the problem of moving target detection is not suitable for real-time applications,an algorithm of ground vehicle target detection was presented by combining unsupervised feature learning and saliency detection. Local features of vehicle targets were learned and encoded to realize saliency detection of the entire image and acquire the potential target area. The highly relevant features can be removed by correlation analysis to effectively suppress the backdrops and highlight the salient targets.
出处
《河南科技大学学报（自然科学版）》
CAS
北大核心
2017年第1期48-51,6,共4页
Journal of Henan University of Science And Technology:Natural Science
基金
国家自然科学基金项目(61401526)
河南省自然科学基金项目(152300410134)
河南省高等学校重点科研软科学计划基金项目(15A880023)
关键词
特征学习
视觉字典
特征提取
feature learning
visual dictionary
feature extraction
