Around view monitor (AVM) based parking slot detection (PSD) is an important task for autonomous parking. Among various algorithms proposed for this task, such as parking line detection or segmentation-based algorithms, corner point detection-based algorithms achieve higher precision since they detect the most fundamental feature of a parking slot. However, such algorithms cannot detect the parking slot when it is partially occluded by the vehicle: even a slight occlusion of one or two corner points significantly disrupts the detection. Moreover, they do not distinguish apart different parking slots when multiple slots exist. To address these problems, this paper proposes a parking slot detection and tracking (PSDT) algorithm using deep neural network, namely PSDT-Net. In order to accurately infer the parking slot position partially occluded in the camera’s blind spot, PSDT-Net detects not only the visible corner points but additional marking points near the blind spot. In addition, using DeepSORT algorithm, PSDT-Net assigns unique labels to different parking slots and tracks their location over time: initially targeted parking slot can be consistently tracked throughout the parking process. Real vehicle experiments show that PSDT-Net can detect and track the parking slot precisely during parking, compared to existing corner point detection-based PSD algorithms.
Deep Learning Based Parking Slot Detection and Tracking: PSDT-Net
Lect. Notes in Networks, Syst.
International Conference on Robot Intelligence Technology and Applications ; 2021 ; Daejeon, Korea (Republic of) December 16, 2021 - December 17, 2021
2022-04-01
12 pages
Article/Chapter (Book)
Electronic Resource
English
Sensor Fusion-Based Vacant Parking Slot Detection and Tracking
Online Contents | 2014
|LiDAR-Based High-Accuracy Parking Slot Search, Detection, and Tracking
SAE Technical Papers | 2020
|LiDAR-Based High-Accuracy Parking Slot Search, Detection, and Tracking
British Library Conference Proceedings | 2020
|