The Need

High-accuracy 3D depth perception is critical for autonomous systems, but current stereo camera methods falter in complex outdoor environments. While LiDAR can improve accuracy, its effectiveness plummets when using sparse data from affordable, lower-beam sensors, making high performance prohibitively expensive. This creates a significant market need for a solution that can fuse sparse, low-cost LiDAR with stereo vision to achieve robust, dense depth maps without the high cost of high-end sensors.

The Technology

OSU engineers have developed Guided RAFT-Stereo (GRAFT-Stereo), a novel software solution that dramatically enhances stereo vision with sparse LiDAR data. Our core innovation is a "depth pre-filling" technique that intelligently interpolates sparse LiDAR points before they are fed into the depth estimation model. This approach overcomes the limitations of current fusion methods by ensuring that even a few hundred LiDAR points can effectively guide the stereo matching algorithm, resulting in a highly accurate and dense 3D depth map.

Commercial Applications

• Automotive: Advanced driver-assistance systems (ADAS) and autonomous vehicles.
• Robotics: Navigation and object interaction for agricultural, logistics, and industrial robots.
• Drones: Autonomous navigation, surveying, and inspection.
• Augmented Reality: Real-world environment mapping.

Benefits/Advantages

• Reduced Cost: Enables high-end performance using inexpensive, low-beam LiDAR sensors.
• Superior Accuracy: Outperforms state-of-the-art methods, especially in sparse LiDAR conditions.
• Seamless Integration: A simple software enhancement to the efficient and flexible RAFT-Stereo framework.