Infrared vehicle detection based on visual saliency and target confidence

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.