Adaptive point cloud acquisition and upsampling for automotive lidar

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Technology

Background:

• Autonomous vehicles require accurate environmental perception for safe operation.
• Lidar sensors offer high range, accuracy, and robustness but often lack sufficient spatial resolution.
• Existing sensor fusion and point cloud processing methods face limitations in detail acquisition.

Purpose of the Study:

• To enhance the spatial resolution of lidar data for autonomous driving applications.
• To develop an adaptive scanning strategy that optimizes point density for object detection.
• To improve the quality of point cloud representations without increasing overall data load.

Main Methods:

• An adaptive paradigm for lidar scanning, prioritizing higher resolution for objects of interest.
• Integration with an auxiliary camera and object detector for initial region proposals.
• Comparison of the proposed adaptive sampling with regular sampling techniques for point cloud upsampling.

Main Results:

• The adaptive strategy improves the quality of object representation in lidar data.
• It achieves better upsampled point cloud quality compared to regular sampling methods.
• The method effectively balances object detail with background data density.

Conclusions:

• Adaptive lidar scanning is a viable strategy to overcome spatial resolution limitations.
• This approach enhances sensor data quality for improved autonomous vehicle perception.
• Optimized point utilization leads to more robust environmental sensing.