Hierarchical Paired Channel Fusion Network for Street Scene Change Detection

Street Scene Change Detection (SSCD) aims to locate the changed regions between a given street-view image pair captured at different times, which is an important yet challenging task in computer vision community. The intuitive way solve SSCD fuse extracted feature pairs, and then directly measure dissimilarity parts for producing change map. Therefore, key design effective fusion method that can improve accuracy of corresponding maps. To this end, we present novel Hierarchical Paired Channel Fusion Network (HPCFNet), utilizes adaptive paired channels. Specifically, features are jointly by Siamese Convolutional Neural (SCNN) hierarchically combined exploring channel pairs multiple levels. In addition, based on observation distribution scene changes diverse, further propose Multi-Part Feature Learning (MPFL) strategy detect diverse changes. Based MPFL strategy, our framework achieves approach adapt scale location diversities regions. Extensive experiments three public datasets (i.e., PCD, VL-CMU-CD CDnet2014) demonstrate proposed superior performance outperforms other state-of-the-art methods with considerable margin.