Fisheye Lens Camera based Autonomous Valet Parking System

TL;DR

This paper presents an autonomous valet parking system using fisheye cameras and visual SLAM, converting fisheye images into surround views for efficient, real-time parking with high detection and success rates.

Contribution

It introduces a novel approach combining fisheye cameras with image conversion and lookup tables for real-time autonomous parking.

Findings

High detection rate of parking environment features

Successful autonomous parking in experimental tests

Real-time performance achieved with minimal computational load

Abstract

This paper proposes an efficient autonomous valet parking system utilizing only cameras which are the most widely used sensor. To capture more information instantaneously and respond rapidly to changes in the surrounding environment, fisheye cameras which have a wider angle of view compared to pinhole cameras are used. Accordingly, visual simultaneous localization and mapping is used to identify the layout of the parking lot and track the location of the vehicle. In addition, the input image frames are converted into around view monitor images to resolve the distortion of fisheye lens because the algorithm to detect edges are supposed to be applied to images taken with pinhole cameras. The proposed system adopts a look up table for real time operation by minimizing the computational complexity encountered when processing AVM images. The detection rate of each process and the success rate of autonomous parking were measured to evaluate performance. The experimental results confirm that autonomous parking can be achieved using only visual sensors.