Optimal Densely Connected Networks with Pyramid Spatial Matching Scheme for Visual Place Recognition

Visual place recognition (VPR) is the procedure of identifying a formerly visited place by the use of visual details under distinct appearance conditions and viewpoint changes. With the recent advances in imaging technologies, camera hardware, and deep learning (DL) models, the VPR becomes a hot research topic among computer vision and robotics communities. The VPR is found to be a challenging process, particularly in uncontrolled outdoor environments and over long duration owing to the environmental factors, light variations, and geometric aspects. With this motivation, this paper presents optimal densely connected network with spatial pyramid matching scheme for VPR, named ODNPSM-VPR technique. The ODNPSM-VPR technique mainly aims to achieve maximum detection performance under varying viewpoints and conditions. The VPR can be considered as an image retrieval process by the inclusion of two stages namely retrieval of closest candidate images and re-ranking. The ODNPSM-VPR technique employs DenseNet-121 model as a feature extractor and the hyperparameters of the DenseNet-121 model are tuned by the grasshopper optimization algorithm (GOA). Besides, image filtering database (II-D) with Manhattan distance is used for the retrieval of top candidate images and pyramid spatial matching scheme is employed for comparing the query image with the candidate images to recognize the place. In order to demonstrate the performance of the proposed ODNPSM-VPR technique, a series of simulations were carried out. The experimental results reported the promising performance of the ODNPSM-VPR technique interms of different measures.