Graph Construction with Soft Fusion of Temporal-Spatial Information for Pedestrian Re-identification

As shown in Figure 1 and Figure 2, the lack of real temporal and spatial labels and the irregularity of pedestrian temporal-spatial transfer often result in unreliable temporal-spatial scores. Therefore, the hard fusion method for integrating temporal-spatial information cannot accurately judge whether the images belong to the same pedestrian or not. The key is to reduce the noise caused by the fusion of temporal-spatial information. Based on this, we propose a method of constructing a graph by soft fusion of temporal-spatial information. By using temporal-spatial information to screen out nodes with higher similarity within a higher range of visual similarity, we construct a graph. The representation of the graph is as follows.