Graph Cut-Based Human Body Segmentation in Color Images Using Skeleton Information from the Depth Sensor.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Human body segmentation is crucial for applications like background substitution, activity recognition, security, and video surveillance.
• Challenges in human body segmentation arise from the complex, non-rigid shapes and motion of the human body.
• Depth sensors offer real-time human skeleton data with notable accuracy, presenting an opportunity for improved segmentation.

Purpose of the Study:

• To propose a novel algorithm for segmenting human bodies in color images.
• To leverage projected human body skeletons from depth images as a primary cue for segmentation.
• To enhance the accuracy and quality of human body segmentation in visual data.

Main Methods:

• Developing an algorithm to project human body skeletons from depth image data onto corresponding color images.
• Utilizing the projected skeleton data as a segmentation cue within the color image.
• Conducting experiments using the Kinect sensor to validate the proposed method.

Main Results:

• The proposed algorithm successfully segments human body regions in color images.
• Experimental results demonstrate high-quality segmentation outcomes.
• The method shows superior performance compared to conventional segmentation techniques.

Conclusions:

• The integration of skeleton data from depth sensors significantly enhances human body segmentation in color images.
• The proposed projection-based method offers a robust and accurate solution for various computer vision applications.
• This approach effectively addresses the challenges posed by non-rigid human body shapes and motion.