Toward featureless visual navigation: Simultaneous localization and planar surface extraction using motion vectors in video streams

Abstract

Abstract-Unlike the traditional feature-based methods, we propose using motion vectors (MVs) from video streams as inputs for visual navigation. Although MVs are very noisy and with low spatial resolution, MVs do possess high temporal resolution which means it is possible to merge MVs from different frames to improve signal quality. Homography filtering and MV thresholding are proposed to further improve MV quality so that we can establish plane observations from MVs. We propose an extended Kalman filter (EKF) based approach to simultaneously track robot motion and planes. We formally model error propagation of MVs and derive variance of the merged MVs. We have implemented the proposed method and tested it in physical experiments. Results show that the system is capable of performing robot localization and plane mapping with a relative trajectory error of less than 5.1%.

Featureless Motion Vector-Based Simultaneous Localization, Planar Surface Extraction, and Moving Obstacle Tracking

“…Since stationary planes are segmented as observations, the problem is reduced to the same problem in [15]. We can employ the same EKF design in [15]. Moving Object Tracking: Similarly, this step is also handled using EKF (the bottom part of Fig.…”

“…To evaluate plane mapping accuracy, we compare our method with our previous work [15] which is referred as SLAPSE method since it only performs localization and plane mapping without ability of tracking moving objects. We use the dataset from [15] for comparison where ground truth is computed by points measured using a laser distance measurer with ±1 mm accuracy.…”

“…With a monocular camera as the only sensor, we have employed MVs from video streams to create a new featureless SLAM method for visual navigation [15]. However, the method assumes a stationary environment despite that MVs encode motion information for both the environment and moving objects.…”

“…Let us briefly introduce MVs here. Detailed description and the filtering process can be found in [15]. …”

“…However, MVs are sparser in spatial resolution but denser in temporal dimension. In [15], we have showed how to exploit this characteristic to reduce noise in MVs, which results in the filtered MVs. Actually, filtered MVs represent the set of corresponding MBs between key frames k and k − 1, and are denoted by…”