Cascaded superpixel pedestrian object segmentation algorithm

Abstract

For the human-being's body contour distortion problem of the pedestrian object segmentation at the complex indoor and outdoor environment for mobile robot visual applications, a cascaded superpixel pedestrian object segmentation algorithm was proposed for considering background interference. Based on acquiring the global superpixel blocks with the primary superpixel computation, the secondary superpixel achieved the correlation degree of the average color and center point Euclidean distance of each superpixel blocks between inside and outside of the pedestrian saliency detection region, in order to obtain the segmentation of the upright person. With the simulation results, this proposed algorithm is 0.9797 in precision-recall statistical average and has excellent target extraction performance compared to state-of-the-art saliency object segmentation algorithm, so that this algorithm can provide the support for the pedestrian object tracking and autonomous driving applications.