A Survey of Deep Learning-Based 3D Object Detection Methods for Autonomous Driving Across Different Sensor...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Autonomous driving systems rely on accurate 3D object detection.
• Deep learning has become a dominant approach for this task.
• Various sensor modalities (cameras, LiDAR, radar) are employed, each with strengths and weaknesses.

Purpose of the Study:

• To provide a comprehensive survey of deep learning-based 3D object detection methods for autonomous driving.
• To systematically categorize these methods based on input sensor modalities.
• To analyze the chronological evolution and quantitative performance of different approaches.

Main Methods:

• A structured taxonomy categorizing methods by input modality (monocular, stereo, LiDAR, radar, fusion).
• Chronological analysis of architectural developments and paradigm shifts.
• Quantitative comparison of surveyed methods using standard metrics on benchmark datasets.

Main Results:

• A modality-agnostic overview of the current state-of-the-art in deep learning for 3D object detection.
• Identification of key trends and performance benchmarks across different sensor inputs.
• A publicly available GitHub repository with survey results.

Conclusions:

• Deep learning approaches for 3D object detection in autonomous driving are diverse and rapidly evolving.
• Sensor modality plays a crucial role in method design and performance.
• This survey offers a valuable resource for researchers and developers in the field.