Robust Estimation and Optimized Transmission of 3D Feature Points for Computer Vision on Mobile Communication Network

Area of Science:

• Computer Vision
• Robotics
• Mobile Computing

Background:

• Multimedia services require efficient data transfer from mobile devices to cloud servers.
• Transmitting raw images raises concerns about data volume and information security.
• Existing methods may not be optimal for distributed computing environments using mobile RGB cameras.

Purpose of the Study:

• To propose a novel method for generating 3D keypoints using a mobile device's RGB camera.
• To enable direct transfer of 3D keypoint information to cloud servers, bypassing raw image transmission.
• To enhance security and efficiency in multimedia data transfer.

Main Methods:

• Capturing images with a moving camera and extracting 2D keypoints.
• Calculating disparities between frames using matched keypoints and estimating baseline distance from camera motion.
• Generating 3D keypoints by combining 2D keypoints with calculated distances.
• Implementing a keypoint-based scene change detection to transfer only new keypoints.

Main Results:

• The proposed method achieved an average error of 5.98 mm compared to the TUM dataset ground truth.
• A significant reduction in transferred 3D keypoints by approximately 73.6% was observed.
• The method demonstrates good performance using a commercial RGB camera on a mobile device.

Conclusions:

• The developed method effectively generates accurate 3D keypoints from mobile RGB cameras.
• The approach significantly reduces data transmission, enhancing security and efficiency for cloud-based multimedia services.
• This technique offers a practical solution for distributed computing environments leveraging mobile sensing capabilities.