RoadTrack: Real-Time Distant Parking Detection via Visual Geometric Projection and Velocity Estimation

📖 Overview

This project provides the first publicly available distant highway abnormal parking event dataset and baseline implementation. The dataset contains 1,000 video sequences, including 200 abnormal parking event videos and 800 normal traffic videos, providing a reliable training foundation and performance evaluation benchmark for highway abnormal event detection research.

🚀 Key Contributions

Highway Abnormal Parking Event Dataset

• Dataset Scale: 1,000 video sequences
• Data Distribution: 200 abnormal event videos + 800 normal traffic videos
• Annotation Quality: Provides detailed object detection and tracking annotations
• Research Value: Provides standardized evaluation benchmarks for subsequent research

🎬 Visualization

The following GIF animations demonstrate the detection process of vehicles transitioning from motion to stationary states at long distances. Green bounding boxes indicate moving vehicles, while red bounding boxes highlight stationary (parked) vehicles that are detected as abnormal parking events.

Event 1	Event 2

Event 3	Event 4

• 🚗 Green Boxes: Moving vehicles being tracked
• 🚨 Red Boxes: Stationary vehicles detected as abnormal parking events
• 📏 Long-distance Detection: Effective detection capability beyond 100 meters
• 🎯 Real-time Tracking: Continuous monitoring of vehicle status changes

📊 Dataset Comparison

The following table presents a comparison of RoadTrack dataset with other existing datasets:

Dataset	Number of Videos	Number of Events	Long Distance
HighD	60	N/A	×
ISLab-PVD	16	22	×
DAD	668	620	×
RoadTrack	1000	300	√

Our Advantages

• 📈 Largest Scale: 1000 video sequences, providing comprehensive coverage
• 🎯 Event Focus: 300 abnormal parking events with detailed annotations
• 🔭 Long-distance Capability: Unique focus on long-distance detection (>100m)

📁 Dataset Structure

datasets/
├── P/                          # Abnormal parking event videos
│   ├── 04-12_09_K221_1953/     # Video sequence (date_time_road_frame format)
│   │   └── data/
│   │       ├── det/            # Object detection results
│   │       └── gt/             # Ground truth annotations
│   ├── ...                     # More abnormal event sequences
├── N/                          # Normal traffic videos
│   ├── 04-12_09_K221_1953/     # Video sequence (date_time_road_frame format)
│   │   └── data/
│   │       ├── det/            # Object detection results
│   │       └── gt/             # Ground truth annotations
│   ├── ...                     # More normal traffic sequences

Video Data Access: To download the full RoadTrack video dataset, please apply for access via the following link: Download RoadTrack Video Data

Data Format Description

• Detection File Format (det/): frame_id,track_id,x1,y1,w,h,conf,class_id
• Ground Truth Format (gt/): frame_id,track_id,x1,y1,w,h,conf,class_id
• Coordinate System: Image coordinate system with top-left as origin, (x1,y1) is top-left corner of bounding box, w,h are width and height

🛠️ Environment Setup

System Requirements

• Python 3.7+
• OpenCV
• NumPy
• SciPy

Install Dependencies

pip install -r requirements.txt

Dependency List

numpy
filterpy
lap
scipy
argparse
opencv-python

🚀 Quick Start

1. Run Baseline Method

python run_baseline.py

2. Run RoadTrack Method

python run_roadtrack.py

3. Run YOLO Video Detection Demo

We provide a demo script in the yolo_detector folder for object detection in videos using the YOLO model. The provided yolo11m.pt is the weight file used in our paper, and video.mp4 is a sample test video from our dataset. For access to the complete video dataset, authorization is required.

Usage:

1. The yolo_detector directory already contains:

   • yolo11m.pt: YOLO model weights used in our publication
   • video.mp4: A sample test video from the RoadTrack dataset

2. Run the demo script:

   cd yolo_detector
   python run_yolo_demo.py

The script will process the sample video and display detection results in real time using OpenCV. You can press q to exit visualization.

📊 Performance Evaluation

Evaluation Metrics

• Detection Precision: Abnormal event detection precision
• Detection Recall: Abnormal event detection recall
• F1: Abnormal event detection F1 Score
• Detection Latency: Abnormal event detection latency

Baseline Performance

Baseline method performance on test set:

• Long-distance parking detection capability: >100m
• Real-time processing capability: Supports real-time video stream analysis

🎯 Applications

• Intelligent Transportation Systems: Real-time monitoring of highway abnormal events
• Traffic Safety Management: Automatic detection and warning of parking events
• Academic Research: Multi-object tracking and abnormal detection algorithm research
• Engineering Applications: Traffic monitoring system development and deployment

📖 Citation

If you use this dataset or method in your research, please cite:

📧 Contact

For questions or suggestions, please contact us through:

• Email: [corfyi@csust.edu.cn]

🙏 Acknowledgments

Thanks to all researchers and developers who contributed to dataset annotation and algorithm development.

---

Keywords: Highway Monitoring, Abnormal Parking Detection, Multi-Object Tracking, RoadTrack, Computer Vision, Intelligent Transportation Systems