Pedestrian Segmentation in Infrared Images Based on Circular Shortest Path

A novel infrared pedestrian segmentation algorithm based on the circular shortest path is proposed. The foreground containing pedestrians is estimated by saliency mapping and gray thresholding. In the foreground area, the human shape feature is introduced by a regional polar transformation. By adding the human shape coefficient to the object term of the cost function, the extracted contour can fit the human shape well while excluding most false alarms. The proposed algorithm performs better in areas with low contrast, and obtains good segmentation quantitatively and qualitatively.