Occlusion handling in parking space detection system
Keywords
No Thumbnail Available
Authors
Issue Date
2020-09-01
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Optimised use of limited resources is the essence of any smart solutions. Optimising parking space usage is
a small subset of this generalised problem. Car users waste a significant share of their prime working hours
looking for an available parking spot. The solutions regarding the tracking and sharing of the availability
of the parking spots have been addressed using a variety of detection methodology. Traditional methods
used electromagnetic sensors which gradually evolved into computer vision-based detection systems. The
robustness of the computer vision-based detection system is challenged under occlusion, and this study
is an attempt to alleviate this flaw. This study proposes a deep convolution neural network-based parking
space detection system capable of tackling common occlusion scenarios in the parking lot. The proposed
model uses mobilenet version 2 as a backbone network of the detection model architecture and Yolo
version 2 as a dynamic occlusion locator integrated into parallel detection process. This study compares
the detection performance of the proposed model with two benchmark detection models (mAlexnet and
VGG-F) in three types of parking scenarios: Unoccluded images scenarios, artificially induced partially
occluded images scenarios and video footage from a real-parking lot. The proposed detection model
outperforms both benchmark detection models in all of the scenarios.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen