Prediction-based perspective warping of feature template for improved visual SLAM accuracy

Abstract

The paper presents an improved method for feature matching in the visual simultaneous localization and mapping (SLAM) system. The appearance of the point feature’s neighborhood observed from a different camera pose is estimated according to the predicted displacement of the camera. As a result the precision of feature matching increases and so does the accuracy of the trajectory’s reconstruction. The proposed method was compared with the state-of-the-art feature detectors and descriptors in the context of visual SLAM. The obtained results place it on par with the best feature descriptors in terms of the system’s accuracy while having significantly smaller computational requirements.