Matching Image Sequences using Mathematical Programming: Visual Localization Applications

This paper proposes a new visual localization algorithm that utilizes the visual route map to localize the agent. The sequence of the current and past images is matched to the map, i.e. the reference image sequence, to produce the best match of the current image. The image sequence matching is achieved by measuring the similarity between the two image sequences using the dynamic time warping (DTW) algorithm. The DTW algorithm employs Dynamic Programming (DP) to calculate the distance (the cost function) between the two image sequences. Consequently, the output of the alignment process is an optimal match of each image in the current image sequence to an image in the reference one. Our proposed DTW matching algorithm is suitable to be used with a wide variety of engineered features, they are SIFT, HOG, LDP in particular. The proposed DTW algorithm is compared to other recognition algorithms like Support Vector Machine (SVM) and Binary- appearance Loop-closure (ABLE) algorithm. The datasets used in the experiments are challenging and benchmarks, they are commonly used in the literature of the visual localization. These datasets are the” Garden point”, “St. Lucia”, and “Nordland”. The experimental observations have proven that the proposed technique can significantly improve the performance of all the used descriptors, i.e, SIFT, HOG, and LDB as compared to its individual performance. In addition, it was able to the SVM and ABLE localization algorithm.