Abstract:Recognizing places from an opposing viewpoint during a return trip is a common experience for human drivers. However, the analogous robotics capability, visual place recognition (VPR) with limited field of view cameras under 180 degree rotations, has proven to be challenging to achieve. To address this problem, this paper presents Same Place Opposing Trajectory (SPOT), a technique for opposing viewpoint VPR that relies exclusively on structure estimated through stereo visual odometry (VO). The method extends recent advances in lidar descriptors and utilizes a novel double (similar and opposing) distance matrix sequence matching method. We evaluate SPOT on a publicly available dataset with 6.7-7.6 km routes driven in similar and opposing directions under various lighting conditions. The proposed algorithm demonstrates remarkable improvement over the state-of-the-art, achieving up to 91.7% recall at 100% precision in opposing viewpoint cases, while requiring less storage than all baselines tested and running faster than all but one. Moreover, the proposed method assumes no a priori knowledge of whether the viewpoint is similar or opposing, and also demonstrates competitive performance in similar viewpoint cases.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is in robot visual place recognition (VPR), how to achieve place recognition of the opposite view (i.e., after a 180 - degree rotation) when using a camera with a limited field - of - view (FOV). This problem is very important in practical applications because it is a common requirement for robots or autonomous vehicles to recognize previously visited places from the opposite direction on the return path. However, unlike human drivers who can easily handle this situation, the existing robotic technologies have been performing unsatisfactorily in this regard, especially in cases where there are large changes in lighting and weather conditions. To address this challenge, the paper proposes a technology named SPOT (Same Place Opposing Trajectory). SPOT relies entirely on the structure estimated by stereo visual odometry (VO) for place recognition in the opposite view. This method extends the recent progress in LiDAR descriptors and introduces a new double - distance - matrix sequence - matching method. This method not only performs well in the opposite view but also outperforms the existing benchmark methods in terms of storage requirements and running speed. Specifically, the main contributions of SPOT include: 1. **Proposing a point - cloud - based method**: Constructing place descriptors from the point cloud generated by stereo visual odometry instead of relying on the field - of - view of a single image. 2. **Developing a new distance - metric method**: By calculating the distance between the query descriptor and the reference descriptor, a variable - offset double - descriptor distance - calculation method is adopted, which improves the robustness to longitudinal and lateral displacements. 3. **Implementing an efficient sequence - matching algorithm**: By performing sequence - matching in two independent distance matrices respectively, effectively identifying place matches in similar and opposite views without prior knowledge of whether the views are the same or opposite. The paper verifies the effectiveness of SPOT through experiments on public datasets, especially the recognition performance in the opposite view is significantly better than existing methods. These results indicate that SPOT has great potential in handling place - recognition problems in complex scenarios such as changes in lighting and weather conditions.

SPOT: Point Cloud Based Stereo Visual Place Recognition for Similar and Opposing Viewpoints

Design of an Enhanced Visual Odometry by Building and Matching Compressive Panoramic Landmarks Online

PALVO: Visual Odometry Based on Panoramic Annular Lens.

A Panoramic Localizer Based on Coarse-to-Fine Descriptors for Navigation Assistance

Visual Place Recognition for Opposite Viewpoints and Environment Changes

SVO-PL: Stereo Visual Odometry with Fusion of Points and Line Segments

RLP‐VIO: Robust and Lightweight Plane‐based Visual‐inertial Odometry for Augmented Reality

Landmark Assisted Stereo Visual Odometry

Stereo Orientation Prior for UAV Robust and Accurate Visual Odometry

DynPL-SVO: A Robust Stereo Visual Odometry for Dynamic Scenes

DynPL-SVO: A New Method Using Point and Line Features for Stereo Visual Odometry in Dynamic Scenes

OCC-VO: Dense Mapping via 3D Occupancy-Based Visual Odometry for Autonomous Driving

Stereo Visual Odometry with Deep Learning-Based Point and Line Feature Matching using an Attention Graph Neural Network

Don't Look Back: Robustifying Place Categorization for Viewpoint- and Condition-Invariant Place Recognition

C2L-PR: Cross-modal Camera-to-LiDAR Place Recognition Via Modality Alignment and Orientation Voting

Human Visual Attention Mechanism-Inspired Point-and-Line Stereo Visual Odometry for Environments with Uneven Distributed Features

PanoVPR: Towards Unified Perspective-to-Equirectangular Visual Place Recognition via Sliding Windows across the Panoramic View

REAL-TIME SLAM FOR THE OFF-ROAD AUTONOMOUS DRIVING

Probabilistic Appearance-Invariant Topometric Localization With New Place Awareness

Are State-of-the-art Visual Place Recognition Techniques any Good for Aerial Robotics?

SVO: Fast semi-direct monocular visual odometry