A Drone-based Approach to Monocular-Camera Distance Estimation using Computer Vision: A Tool for Surround View System Validation

Abstract

Autonomous parking is a field within ADAS/AD that requires exceptionally high spatial precision, given the close proximity to vehicles, people, and other obstacles. Surround-view camera systems are often used to both map the environment and estimate the vehicle’s position by creating a stitched birds-eye view image of the vehicle’s surroundings. To aid the development and evaluation of these systems, high-accuracy distance measurements between the car and its environment are needed. This thesis presents a drone-based system that utilizes a monocular camera to estimate ground-level distances in parking environments. Using car-mounted AR fiducial markers as reference objects, the system achieves centimeter-level accuracy at distances up to 10 meters relative to the car. In addition to the distance estimation pipeline, a vision-based drone controller was implemented to track a target car and maintain optimal positioning for video data collection during dynamic parking scenarios. Real-world testing demonstrates that the proposed method yields consistent results, highlighting its potential as a low-cost and flexible tool for generating ground truth distances. To our knowledge, this is the first application of monocular drone-based distance estimation for the development and validation of ADAS/AD systems.