Enhancing simultaneous localization and mapping with low- and high- level features

Abstract

In this thesis, we aim to enhance Simultaneous Localization and Mapping (SLAM) with low- and high- level features. The low-level features - usually obtained by keypoint detection - are often clustered, redundant, and noisy. These keypoints usually require special processing like Adaptive Non-Maximal Suppression(ANMS) to keep the most relevant ones. To handle this issue, we present three new efficient ANMS approaches which ensure a fast and homogeneous repartition of the keypoints in the image. Furthermore, we present a robust approach for road marking detection and recognition from the images captured by a camera located inside a vehicle. These road markings are considered as high-level features which can be used to support and improve SLAM in many ways. Therefore, we propose to use road markings to efficiently detect loops in order to correct the accumulated error during localization and mapping. We have performed an extensive number of experiments which highlight the effectiveness and scalability of proposed methods.