NeoARCADE: Robust Calibration for Distance Estimation to Support Assistive Drones for the Visually Impaired

TL;DR

NeoARCADE tackles the challenge of estimating absolute distances from monocular drone imagery for VIP navigation by introducing Neo, a depth-map calibration pipeline that uses depth-score normalization and dynamic recalibration. It pairs Neo with two lightweight baselines, Regression and Geometric, to establish strong comparisons and demonstrates substantial accuracy gains over SOTA monocular-depth methods across diverse outdoor scenes and adversarial conditions. The results show VIP-distance errors typically within $±30$ cm and obstacle distances within $±60$ cm, enabling safer obstacle avoidance and reliable following behavior on edge hardware. The work provides extensive campus datasets, analyzes robustness and generalization across VIPs and environments, and outlines future work toward integrating distance estimation with local path planning for complete assistive navigation.

Abstract

Autonomous navigation by drones using onboard sensors, combined with deep learning and computer vision algorithms, is impacting a number of domains. We examine the use of drones to autonomously follow and assist Visually Impaired People (VIPs) in navigating urban environments. Estimating the absolute distance between the drone and the VIP, and to nearby objects, is essential to design obstacle avoidance algorithms. Here, we present NeoARCADE (Neo), which uses depth maps over monocular video feeds, common in consumer drones, to estimate absolute distances to the VIP and obstacles. Neo proposes robust calibration technique based on depth score normalization and coefficient estimations to translate relative distances from depth map to absolute ones. It further develops a dynamic recalibration method that can adapt to changing scenarios. We also develop two baseline models, Regression and Geometric, and compare Neo with SOTA depth map approaches and the baselines. We provide detailed evaluations to validate their robustness and generalizability for distance estimation to VIPs and other obstacles in diverse and dynamic conditions, using datasets collected in a campus environment. Neo predicts distances to VIP with an error <30cm, and to different obstacles like cars and bicycles within a maximum error of 60cm, which are better than the baselines. Neo also clearly out-performs SOTA depth map methods, reporting errors up to 5.3-14.6x lower.

Figures (20)

• Figure 1: Distance estimation using proposed methods. True distance to VIP is $3m$, bystander is $4m$, car is $4.6m$.
• Figure 2: Comparison of Neo and Geo* with SOTA depth map approaches for distance estimation.
• Figure 3: Illustration of different distance components involved in our distance estimation model.
• Figure 4: Workflow for distance estimation using Regression-based Model.
• Figure 5: Train/Test Dataset for Regression Model.