Computer-Aided Depth Video Stream Masking Framework for Human Body Segmentation in Depth Sensor Images

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Biomedical Imaging

Background:

• Human activity identification from videos is crucial for various applications.
• Three-dimensional (3D) depth images provide spatial positioning information from depth sensors.
• Manual annotation of depth images for segmentation is time-consuming.

Purpose of the Study:

• To present a framework for creating foreground-background masks from depth images for human body segmentation.
• To accelerate the manual depth image annotation process.
• To enable efficient segmentation with minimal user input.

Main Methods:

• Developed a framework for foreground-background mask generation from 3D depth images.
• Implemented a performant segmentation algorithm.
• Allowed user interaction for parameter adjustment, result correction, or boundary hints.

Main Results:

• The framework successfully segmented human bodies in depth images.
• Demonstrated significant reduction in manual segmentation time (both processing and human input).
• Achieved promising results on two real-world datasets.

Conclusions:

• The proposed framework effectively speeds up depth image annotation for human body segmentation.
• It offers an efficient solution for preparing datasets for human activity recognition.
• The approach balances automated segmentation with necessary user guidance.