Augmenting Loop Detection Performance in Autonomous Navigation: A Temporal Similarity-Based Approach

Loop detection constitutes an indispensable component in robotics and autonomous driving navigation systems. Extant literature has predominantly focused on extracting essential features from sensor data and constructing robust descriptors. Nevertheless, these measurements frequently suffer from errors, attributable to the diversity in traversal behaviors, complex traffic environments, and inconsistencies in sensor performance. This paper explores into the spatial and temporal aspects of loops, examines the temporal similarity and patterns of consecutive loop sequences, and introduces a temporal similarity assessment for measurements within the same sequence. Contrary to previous methodologies, our proposed method amalgamates measurements and temporal similarity for detection simultaneously. This innovative approach not only improves the efficiency of the measurement method but also utilizes temporal similarity to offset measurement inaccuracies, thereby enhancing detection performance in complex and extreme scenarios. To corroborate the efficacy of the proposed methodology, a meticulous evaluation is executed on a dataset comprising a wide spectrum of traffic scenarios. The results corroborate that the proposed method exhibits adaptability across an extensive array of complex and extreme scenarios while upholding efficiency.