A survey on 3D object detection in real time for autonomous driving

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• 2D object detection has limitations in dynamic environments.
• Autonomous driving requires robust 3D perception for safety.

Purpose of the Study:

• To review state-of-the-art 3D object detection methods for autonomous driving.
• To provide a taxonomy of existing 3D object detection techniques.
• To highlight current trends and future research directions.

Main Methods:

• Review of monocular and stereo vision-based 3D object detection.
• Categorization of methods into model-based, end-to-end, and hybrid approaches.
• Analysis of multi-view detectors and their robustness.

Main Results:

• Identified drawbacks of 2D methods in dynamic scenarios.
• Presented a taxonomy based on depth inference, learning schemes, and internal representation.
• Highlighted multi-view detectors as a robust end-to-end trend.

Conclusions:

• 3D object detection is crucial for autonomous driving systems.
• Diverse datasets and metrics are essential for evaluating performance.
• Future research should address challenges like occlusion using multi-modal data.