Self-Supervised Object Distance Estimation Using a Monocular Camera.

Area of Science:

• Computer Vision
• Machine Learning
• Robotics

Background:

• Monocular camera-based distance estimation is a fundamental computer vision challenge.
• Existing methods often require extensive data or yield suboptimal accuracy.
• There is a need for integrated and precise single-camera distance estimation solutions.

Purpose of the Study:

• To develop an efficient and accurate network for simultaneous object detection and distance estimation using a monocular camera.
• To improve the precision of monocular distance estimation by integrating camera calibration and multi-scale resolution techniques.
• To validate the proposed network's performance and generality across different scenarios.

Main Methods:

• A hybrid network combining ShuffleNet and YOLO for object detection.
• A self-supervised learning network for distance estimation.
• Integration of calibrated camera parameters and analysis of camera pose variations.
• Application of multi-scale resolution to enhance depth information representation.

Main Results:

• The proposed network achieved efficient and accurate object detection and distance estimation on the KITTI dataset.
• Validation on a newly constructed dataset demonstrated the network's generality in diverse scenarios.
• The approach outperformed alternative methods in object-specific distance estimation.

Conclusions:

• The integrated network offers a robust solution for monocular distance estimation and object detection.
• The method demonstrates superior performance and accuracy compared to existing techniques.
• The developed approach shows significant potential for real-world applications requiring precise distance measurements from monocular vision.