广义二维PCA和稀疏表示的红外行人目标检测

红外行人检测在夜间智能视频监控，车辆安全驾驶等领域有重要应用。为了解决红外图像特征降维后空间结构信息丢失的问题，提出一种基于广义二维主分量分析（principal component analysis，PCA）和稀疏表示的红外图像行人目标检测算法。该算法主要由2个阶段组成：第1阶段利用广义二维主分量分析方法提取图像的二维主特征分量，并由此构造行人目标的超完备特征字典；第2阶段采用滑动窗口的方法得到图像中局部子图，然后利用基追踪算法求解每个局部子图的稀疏表示系数向量，最后定义一个函数度量每个子图存在行人目标的可能性，并设置相邻标记框的最小距离得到整幅图像最终的检测结果。实验结果表明，该方法能够有效地检测红外图像中的行人目标，具有较好的检测效果。

Infrared pedestrian detection based on generalized two-dimensional PCA and sparse representation

Infrared pedestrian detection has many important applications in the fields of night intelligent video surveillance, vehicle driving safety and so on. In order to solve the problem of losing image spatial structure information after dimensionality reduction, an algorithm based on the generalized two-dimensional principal component analysis and sparse representation is proposed in this paper. The algorithm mainly consists of two phases: the first phase is that the two-dimensional principal component features are extracted by using generalized two-dimensional principal component analysis and then an over complete feature dictionary of pedestrian targets is constructed. The second phase is that local sub-images are obtained by using a sliding window and then the sparse representation coefficient vectors of each sub-image are achieved by using the Basis Pursuit algorithm. Finally, a function is defined to measure the certainty of a pedestrian exist in the sub images and then the final result of the whole image is achieved by setting minimum distance between the adjacent marked boxes. Experimental results show that the proposed method can effectively detect pedestrians in infrared images and has good performance.