Fusion of range and stereo data for high-resolution scene-modeling

Area of Science:

• Computer Vision
• 3D Reconstruction
• Sensor Fusion

Background:

• Depth map generation is crucial for 3D scene understanding.
• Existing methods struggle with fusing low-resolution depth data and high-resolution stereo data effectively.
• Accurate disparity map generation is essential for high-quality depth maps.

Purpose of the Study:

• To develop a robust and efficient range-stereo fusion technique for creating high-resolution depth maps.
• To improve the accuracy of depth map construction by intelligently combining different sensor data.
• To overcome limitations of existing methods in handling sparse depth information and stereo inaccuracies.

Main Methods:

• A maximum a posteriori (MAP) formulation is used to fuse low-resolution depth data with high-resolution stereo data.
• The method employs hierarchical local energy minimization, growing sparse initial disparities from depth data.
• Key innovations include a subpixel-corrected correlation function, adaptive cost aggregation, and adaptive fusion of stereo and depth likelihoods.

Main Results:

• The proposed method achieves accurate disparity map generation by avoiding propagation of incorrect disparities.
• It demonstrates superior performance compared to state-of-the-art methods in both quantitative and qualitative evaluations.
• The algorithm is efficient, running at 3 frames per second on 2.0 MP images on standard hardware.

Conclusions:

• The novel range-stereo fusion approach significantly enhances depth map resolution and accuracy.
• The adaptive data-term properties ensure a more selective and reliable disparity growing process.
• This method offers an efficient solution for real-time high-resolution depth map generation in computer vision applications.