基于视觉显著性和目标置信度的红外车辆检测技术

Infrared vehicle detection based on visual saliency and target confidence

提出了一种复杂地面背景下的红外车辆检测算法。首先,提出一种新的自适应分段线性灰度拉伸方法来增强当图像整体亮度偏低时的目标信息。其次,利用拉伸后图像的显著性图生成目标潜在的兴趣区。再次,利用平均梯度法在兴趣区内进行目标的边缘再分割,完成目标精确分割检测。最后,利用车辆的红外融合特征计算目标置信度,对目标进行评估和确认。实验结果表明:对实际拍摄的红外图像进行检测的算法可有效地检测出地面车辆目标。

An algorithm of infrared vehicle detection under the complex background was propose. First, a novel self-adaption piecewise linear stretching function was utilized to enhance the targets when the overall intensity of image was low. Second, the saliency map of the enhanced image was used to generate the ROIs of the potential targets. Then, the vehicle targets were able to be detected by the edge-based segmentation in ROIs using average gradient. Finally, multiple features were fused to discriminate the belief of a region belonging to a vehicle target. Experimental results on the real infrared images show that the algorithm can effectively detect the vehicle target on ground.