Sensing Range Extension for Short-Baseline Stereo Camera Using Monocular Depth Estimation.

Area of Science:

• Computer Vision
• Robotics
• Sensor Fusion

Background:

• Short-baseline stereo cameras (SBSCs) have limited sensing ranges.
• Monocular depth estimation (MDE) using deep learning offers potential but often lacks accurate scale.
• Integrating complementary depth data sources is crucial for robust perception.

Purpose of the Study:

• To propose a novel method for extending the sensing range of SBSCs.
• To effectively fuse stereo depth information with MDE.
• To enable robust depth estimation across diverse environments without retraining.

Main Methods:

• A convolutional neural network-based monocular depth estimation (MDE) model is employed.
• Stereo depth data is used to estimate a scale factor for the monocular depth.
• The scaled monocular depth is then combined with the stereo depth.

Main Results:

• The proposed method successfully extends the effective sensing range of SBSCs.
• Quantitative and qualitative evaluations demonstrate the accuracy of the combined depth estimation.
• The MDE model shows adaptability to different environments post-training.

Conclusions:

• The fusion of stereo and scaled monocular depth significantly improves SBSC performance.
• The method offers a practical solution for enhancing depth perception in various applications.
• The approach demonstrates robustness and reusability of trained MDE models.