Vision-based Parking-slot Detection: A DCNN-based Approach and A Large-scale Benchmark Dataset

Area of Science:

• Computer Vision
• Artificial Intelligence
• Automotive Engineering

Background:

• Self-parking systems require accurate parking-slot detection.
• Vision-based detection is challenging due to varied lighting, ground materials, and shadows.

Purpose of the Study:

• To develop an accurate and efficient vision-based parking-slot detection method.
• To create a large-scale dataset for advancing parking-slot detection research.

Main Methods:

• Proposed DeepPS, a Deep Convolutional Neural Network (DCNN) approach using surround-view images.
• Developed a two-step process: identifying marking points and classifying local image patterns.
• Created and labeled a large dataset of 12,165 images.

Main Results:

• DeepPS demonstrated efficacy and efficiency in parking-slot detection.
• The developed dataset is the largest publicly available for this task.
• Source code and dataset are publicly accessible for reproducibility.

Conclusions:

• The DeepPS approach offers a viable solution for challenging parking-slot detection scenarios.
• The new dataset will accelerate research and development in autonomous parking.