Reconstructed Saliency for Infrared Pedestrian Images

Accurately and completely detecting infrared pedestrian is a challenging problem in an intelligent transportation system due to the low SNR and the inhomogeneous luminance distribution in the infrared images, especially for the complex background environment. In this paper, we introduce a reconstruction optimization-based saliency detection method for infrared pedestrian images to solve the problem. First, appearance-based infrared saliency was introduced to enhance the salient areas of the infrared images from locally and globally contrast features. Then, considering the essential characteristic of the infrared pedestrian images, thermal radiation prior, and pedestrian shape prior was combined to construct infrared object prior information. Finally, the infrared pedestrian saliency map can be calculated through a random walk-based saliency reconstruction optimization method with the appearance saliency and infrared object prior. The extensive experiments on real infrared images captured by intelligent transportation systems demonstrate that our saliency algorithm consistently outperforms the state-of-the-art saliency detection methods, in terms of higher precision, F-measure, and lower mean absolute error.