HMD-Net: A Vehicle Hazmat Marker Detection Benchmark.

Area of Science:

• Computer Vision and Machine Learning
• Transportation Safety Systems

Background:

• Vehicles transporting hazardous materials (hazmat) pose significant risks to highway transportation safety.
• Automated recognition of hazmat markers on vehicles is crucial for intelligent transportation management systems.
• A lack of public datasets hinders the development and benchmarking of hazmat marker detection models.

Purpose of the Study:

• To introduce the VisInt-VHM, a novel, large-scale dataset for vehicle hazmat marker detection.
• To develop and evaluate HMD-Net, a compact and efficient deep learning model for hazmat marker detection.
• To provide a benchmark for evaluating hazmat marker detection models in real-world highway conditions.

Main Methods:

• Creation of the VisInt-VHM dataset: 10,000 images with 20,023 hazmat markers captured under diverse environmental conditions.
• Development of HMD-Net: A lightweight convolutional neural network architecture optimized for efficiency.
• Model compression techniques, including channel pruning, applied to HMD-Net for edge device deployment.

Main Results:

• HMD-Net demonstrates a superior trade-off between detection accuracy and inference speed compared to established models like YOLOv3/v4 and other lightweight networks.
• The proposed model achieves efficient performance, suitable for deployment on resource-constrained edge devices.
• The VisInt-VHM dataset provides a valuable resource for advancing research in vehicle hazmat detection.

Conclusions:

• The VisInt-VHM dataset and HMD-Net offer a significant advancement for automated vehicle hazmat marker detection.
• HMD-Net provides an efficient solution for real-time hazmat detection, enhancing transportation safety.
• This work addresses the need for robust and deployable models in intelligent transportation systems.