Robust Monocular Visual Teach and Repeat Aided by Local Ground Planarity and Color‐constant Imagery

Abstract

Visual Teach and Repeat (VT&R) allows an autonomous vehicle to accurately repeat a previously traversed route using only vision sensors. Most VT&R systems rely on natively three‐dimensional (3D) sensors such as stereo cameras for mapping and localization, but many existing mobile robots are equipped with only 2D monocular vision, typically for teleoperation. In this paper, we extend VT&R to the most basic sensor configuration—a single monocular camera. We show that kilometer‐scale route repetition can be achieved with centimeter‐level accuracy by approximating the local ground surface near the vehicle as a plane with some uncertainty. This allows our system to recover absolute scale from the known position and orientation of the camera relative to the vehicle, which simplifies threshold‐based outlier rejection and the estimation and control of lateral path‐tracking error—essential components of high‐accuracy route repetition. We enhance the robustness of our monocular VT&R system to common failure cases through the use of color‐constant imagery, which provides it with a degree of resistance to lighting changes and moving shadows where keypoint matching on standard gray images tends to struggle. Through extensive testing on a combined 30 km of autonomous navigation data collected on multiple vehicles in a variety of highly nonplanar terrestrial and planetary‐analogue environments, we demonstrate that our system is capable of achieving route‐repetition accuracy on par with its stereo counterpart, with only a modest tradeoff in robustness.

Learning Matchable Image Transformations for Long-Term Metric Visual Localization

“…Our goal in this work is to learn a nonlinear transformation f : R 3 → R mapping the RGB colorspace onto a grayscale colorspace that explicitly maximizes a chosen performance metric of a vision-based localization pipeline. We investigate two approaches to formulating such a mapping: 1) a single function to be applied as a pre-processing step to all incoming images, similarly to [11], [13], [14]; and 2) a parametrized function tailored to the specific image pair to be used for localization, where the parameters of this function are derived from the images themselves. Additionally, the functional form of either mapping may be specified analytically (e.g., from physics) or learned from data using a function approximator such as a neural network.…”

“…However, in practice, these constraints are relaxed and the parameters (α, β) are tuned to a specific environment, sensor, and feature matcher, where the theoretical values do not perform optimally. Indeed, [11], [14] used two sets of parameters tuned to maximize the stability of SURF features [32] in regions where grassy or sandy materials dominate.…”

How to Train a CAT: Learning Canonical Appearance Transformations for Direct Visual Localization Under Illumination Change

“…Illumniation robustness in visual localization has been previously studied from the perspective of illumination invariance, with methods such as [12]- [14] making use of hand-crafted image transformations to improve feature matching over time. Similarly, affine models [2] and other analytical transformations [15] have been used to improve the robustness of direct visual localization to illumination change.…”

“…Our direct localization pipeline operates in both mapping (VO) and relocalization modes in a similar vein to topometric visual teach-and-repeat navigation [13], [14], where the camera follows a similar trajectory during both mapping and relocalization phases. As the camera explores the environment in mapping mode, we generate a list of posed keyframes with corresponding image and depth data, creating new keyframes when the translational or rotational distance between the most recent keyframe pose and the current tracking pose exceeds a preset threshold.…”

“…To this end, we investigated the use of our trained CAT models for teach-and-repeat-style metric relocalization [13], [14] by first creating a map in the canonical condition, then relocalizing against it in different conditions using both the original and transformed images. Figure 6 shows sample relocalization errors for the "Morning" and "Sunset" conditions of the VKITTI/0001 trajectory using both the original image streams (resized and cropped to 256 × 192 resolution) and the output of our trained model, while Table II summarizes our relocalization results for all ETHL/syn and VKITTI sequences.…”