Adaptive restoration of pedestrian appearance in the spatiotemporal domain

Abstract

Video synopsis based on visual rhythm has a small amount of calculation and strong representativeness, but the drawback is that the spatiotemporal pedestrians are distorted and illegible due to the loss of their own unique spatial morphology information, which leads to poor performance of video synopsis. To solve this problem, this letter proposed an adaptive restoration method of pedestrian appearance in the spatiotemporal domain. First, a single‐pixel row is taken as the sampling unit to complete the horizontal visual rhythm stitching; second, the relationship between the pedestrian spatiotemporal trajectory slope in the horizontal visual rhythm and the pedestrian movement speed is deduced; finally, the spatiotemporal trajectory slope is used as the basis for the vertical visual rhythm sampling width, and the adaptive restoration of pedestrian appearance in the spatiotemporal domain is completed. Experimental results show that the contour fluency and visual perception effect of pedestrian appearance restored by the proposed method are better, and the difference ratio of the proposed method is 0.08566, and the average running time is 421.0760 s, which has obvious advantages over the comparison method.