The use of local detectors and descriptors in typical computer vision pipelines work well until variations in viewpoint and appearance change become extreme. Past research in this area has typically focused on one of two approaches to this challenge: the use of projections into spaces more suitable for feature matching under extreme viewpoint changes, and attempting to learn features that are inherently more robust to viewpoint change. In this paper we present a novel framework that combines learning of invariant descriptors through data augmentation and orthographic viewpoint projection. We propose rotation-robust local descriptors, learnt through training data augmentation based on rotation homographies, and a correspondence ensemble technique that combines vanilla feature correspondences with those obtained through rotation-robust features. Using a range of benchmark datasets as well as contributing a new bespoke dataset for this research domain, we evaluate the effectiveness of the proposed approach on key tasks including pose estimation and visual place recognition. Our system outperforms a range of baseline and state-of-the-art techniques, including enabling higher levels of place recognition precision across opposing place viewpoints, and achieves practically-useful performance levels even under extreme viewpoint changes.

No abstract

Mobile robots are widely used in the surveillance industry, for military and industrial applications. In order to carry out surveillance tasks like urban search and rescue operation, the ability to traverse stairs is of immense significance. This paper presents a deep learning-based approach for semantic segmentation of stairs, behavioral cloning for stair alignment, and a novel mechanical design for an autonomous stair climbing robot. The main objective is to solve the problem of locomotion over staircases with the proposed implementation. Alignment of a robot with stairs in an image is a traditional problem, and the most recent approaches are centered around hand-crafted texture-based Gabor filters and stair detection techniques. However, we could arrive at a more scalable and robust pipeline for alignment schemes. The proposed deep learning technique eliminates the need for manual tuning of parameters of the edge detector, the Hough accumulator and PID constants. The empirical results and architecture of stair alignment pipeline are demonstrated in this paper.

No abstract