人类视觉机制与ROI融合的红外行人检测

Pedestrian detection in infrared images using ROI fusion and human visual mechanism

为提高行人检测系统在红外场景中的检测率以及速度,提出一种基于人类视觉机制与ROI融合的红外行人检测方法。根据人类视觉机制来改进LoG滤波抑制背景噪声,通过对滤波后的图像应用ROI融合得到行人候选区域,使提取到的ROI更为准确。另外,提出一种改进的纹理特征OCS-LBP(oriented center symmetric local binary patterns),对得到的行人候选区域提取HOG特征和OCS-LBP特征,使用随机蕨分类器来进行分类,提升检测的速度与精度。该方法通过与流行的检测算法比较,检测准确率与召回率分别提升7.9%与10.3%,且实时性有较大的提升,具有一定的研究和实用价值。

In order to improve the robustness and the speed of pedestrian detection system in infrared scene,a pedestrian detection method in infrared image based on human visual mechanism and ROI fusion is proposed.According to the human visual mechanism,improved LoG filtering is used to suppress background noise.And a pedestrian candidate region is obtained by applying ROI fusion to the filtered image.What’s more,an improved low-dimension texture feature OCS-LBP is proposed.The HOG feature and OCS-LBP feature are extracted from the obtained pedestrian candidate region,and finally the random fern classifier is used for classification.Compared with the popular detection algorithm,detection accuracy and recall rate of the proposed method are improved by 7.9%and 10.3%,respectively,and the real-time performance is greatly improved.It has certain research and practical value.