Vision-Based Recognition of Human Motion Intent during Staircase Approaching.

Area of Science:

• Robotics
• Computer Vision
• Human-Computer Interaction

Background:

• Real-world navigation requires continuous decision-making, such as choosing to ascend or avoid stairs.
• Recognizing motion intent for assistive robot control (e.g., prosthetic limbs) is crucial but difficult due to limited information.

Purpose of the Study:

• To develop a novel vision-based method for recognizing an individual's motion intent when approaching a staircase.
• To enable assistive robots to anticipate and adapt to user intentions before a change in motion mode occurs.

Main Methods:

• Utilized egocentric images from a head-mounted camera.
• Trained a YOLOv5 object detection model to identify staircases.
• Developed an AdaBoost and gradient boost (GB) classifier to determine intent (engage or avoid stairs).

Main Results:

• Achieved reliable motion intent recognition with 97.69% accuracy.
• Successfully predicted intent at least two steps before the potential transition to stair climbing.
• Demonstrated the method's effectiveness in providing timely information for robot controller adjustments.

Conclusions:

• The proposed vision-based approach reliably predicts stair-climbing intent using egocentric vision.
• This method offers a significant advancement for assistive robot control, enhancing real-world usability.
• Early intent recognition allows for proactive adjustments in assistive robotic systems.