摘要
针对于稀疏编码在行人检测问题中提取的特征维数高和不能够有效描述行人的问题,提出了一种基于多重稀疏字典直方图的特征提取方法。通过稀疏表示方法,预先学习多个不同稀疏度的字典,分别利用每一个字典对行人图像进行稀疏编码,统计每个字典中对应稀疏编码单元的分布直方图作为行人图像的特征描述子。该方法提取到的特征维数低,并且能够有效地描述行人,具有良好的检测性能。
The drawbacks of pedestrian detection method based on sparse code are high dimensions of features and can’t ef-fectively describe the pedestrian. Aiming at the drawbacks,a feature extraction method based on multiple sparse dictionaries his-togram is proposed. Several different sparse dictionaries need to be learned before hand by means of the sparse representation, sparse coding of pedestrian image is conducted with different sparse dictionaries to make statistics of the distribution histograms corresponding to the sparse coding units in each dictionary as the image feature descriptor. The feature dimensions extracted with this method are low,can effectively describe the pedestrian,and has good detection performance.
出处
《现代电子技术》
北大核心
2015年第2期83-87,共5页
Modern Electronics Technique
基金
国家自然基金"基于粒度空间特征的行人检测方法研究"(61300161)
教育部博士点基金(20133219120033)
江苏省社会安全图像与视频理解重点实验室(南京理工大学)开放基金项目(JSKL201306)
关键词
行人检测
特征提取
稀疏表示
多重稀疏
字典
pedestrian detection
feature extraction
sparse representation
multiple sparse
dictionary
