Visual Odometry by Multi-frame Feature Integration

Abstract

This paper presents a novel stereo-based visual odometry approach that provides state-of-the-art results in real time, both indoors and outdoors. Our proposed method follows the procedure of computing optical flow and stereo disparity to minimize the re-projection error of tracked feature points. However, instead of following the traditional approach of performing this task using only consecutive frames, we propose a novel and computationally inexpensive technique that uses the whole history of the tracked feature points to compute the motion of the camera. In our technique, which we call multi-frame feature integration, the features measured and tracked over all past frames are integrated into a single, improved estimate. An augmented feature set, composed of the improved estimates, is added to the optimization algorithm, improving the accuracy of the computed motion and reducing ego-motion drift. Experimental results show that the proposed approach reduces pose error by up to 65% with a negligible additional computational cost of 3.8%. Furthermore, our algorithm outperforms all other known methods on the KITTI Vision Benchmark data set.

An assembly system based on industrial robot with binocular stereo vision

“…So, it was used to pinpoint the objects [8,9] and the mode is shown in Figure 6. The greatest difference of this model is to bridge the top-down and the bottom-up attention mechanism with 'grouping' such as a point, a feature, an object, a region or a hierarchical structure [10,11]. In this paper, color, intensity and orientation as well as the object contours are added in grouping.…”

Understanding how camera configuration and environmental conditions affect appearance-based localization

“…Details of the algorithm are found in [5] and [6]. Metric localization, such as those based on SLAM [2] and visual odometry [4] are accurate for short distances, but they drift over time or fail at long sequences, especially if the vehicle does not revisit the same place on its trajectory. On the other hand, topological localization methods, such as [23] estimate the observer's location qualitatively from a finite set of possible positions avoiding localization drift but provide only rough position estimates.…”

“…The most used techniques, such as SLAM [9], [13] and visual odometry [4], rely on geometric constraints on the structure of the world and its projection onto the images. These methods usually perform feature tracking, triangulation, and pose estimation, providing metric information of the location of the cameras.…”

Fully Automatic Feature-Based Registration of Mobile Mapping and Aerial Nadir Images for Enabling the Adjustment of Mobile Platform Locations in GNSS-Denied Urban Environments

“…In the research field of autonomous driving, many methods rely on lane marking detection in conjunction with a digital map to support the localisation task (Gruyer et al (2014), Schindler (2013), Roh et al (2016)). A related method is visual odometry where features are tracked across multiple frames of the sensor system installed on the platform to allow for a better relative positioning (Badino et al (2013), Zhang and Singh (2015)). These approaches are all designed to work in a real-time fashion to reduce GNSS-induced localisation issues in difficult scenarios, but cannot reach decimetre-grade accuracy, since external reference information is missing.…”

Learning Visual Odometry with a Convolutional Network

“…In the past few years many interesting and efficient approaches like (Badino et al, 2013;Nistér et al, 2004;Chandraker et al, 2013;Kitt et al, 2010) were introduced to deal with the visual odometry problem. Some of the approaches, like (Chandraker et al, 2013), only rely on monocular video, whereas others, like (Badino et al, 2013;Nistér et al, 2004), use stereo information. A common property across most of these works is that they rely on key-point detection and tracking, combined with camera geometry, for estimating visual odometry.…”