Salience-guided Ground Factor for Robust Localization of Delivery Robots in Complex Urban Environments

TL;DR

This work tackles robust localization for delivery robots in complex urban environments by introducing Salient Ground Features (SGFs) detected from a calibrated BEV through Salient Object Detection. The system leverages Motion Compensate Inverse Perspective Mapping (MC-IPM) to produce stable ground features, a rotation-robust SGF descriptor for loop closure, and a Horn-like 4-step Hu-moment-based selection to identify salient features for matching. Experimental results in campus-like environments show significant improvements in trajectory accuracy over Odometry and ORB-SLAM3, including under night-time conditions, and demonstrate robustness to uneven ground via MC-IPM. The approach offers a low-cost, vision-based localization enhancement by exploiting nonstandard, human-salient ground cues without relying on predefined semantic classes, with potential for integration into broader SLAM frameworks.

Abstract

In urban environments for delivery robots, particularly in areas such as campuses and towns, many custom features defy standard road semantic categorizations. Addressing this challenge, our paper introduces a method leveraging Salient Object Detection (SOD) to extract these unique features, employing them as pivotal factors for enhanced robot loop closure and localization. Traditional geometric feature-based localization is hampered by fluctuating illumination and appearance changes. Our preference for SOD over semantic segmentation sidesteps the intricacies of classifying a myriad of non-standardized urban features. To achieve consistent ground features, the Motion Compensate IPM (MC-IPM) technique is implemented, capitalizing on motion for distortion compensation and subsequently selecting the most pertinent salient ground features through moment computations. For thorough evaluation, we validated the saliency detection and localization performances to the real urban scenarios. Project page: https://sites.google.com/view/salient-ground-feature/home.

Figures (10)

• Figure 1: Illustration of the proposed method. We extract and describe the salient feature on the ground from the BEV. It can be used as a loop factor to perform localization.
• Figure 2: The overall pipeline of the proposed system. The diagram describes utilizing SGF for localization and loop closure.
• Figure 3: Illustration of an MC-IPM model. A corrected BEV image is generated from a compensated 3D point projected through the MC-IPM. The lower right example is a compensation result of a situation crossing a speed bump.
• Figure 4: An example of a SGF detected in a valid feature range. In the above case, two SGFs were detected in the three final groups, and not selected in final group 2 (boundary condition).
• Figure 5: Illustration of SGF description process and rotational invariance through column-shifted matching. The red dot is the center of the SGF points, and the green circle is the radius $L_{max}$ boundary.