Improving the Head Pose Variation Problem in Face Recognition for Mobile Robots.

Area of Science:

• Robotics
• Computer Vision
• Human-Robot Interaction (HRI)

Background:

• Face recognition is crucial for human-robot interaction (HRI) in diverse environments like healthcare and education.
• Extreme head pose variability poses a significant challenge to current face recognition systems in unconstrained settings.
• Existing state-of-the-art methods show decreased effectiveness when faces deviate from limited pose ranges.

Purpose of the Study:

• To address the challenge of extreme head pose variability in robot face recognition.
• To develop and evaluate a novel optimization method for enhancing face recognition robustness.
• To contribute a new dataset and analysis of pose variability effects on recognition systems.

Main Methods:

• Designed a tool to collect a new dataset with a uniform distribution of head pose images.
• Analyzed the performance degradation of state-of-the-art face recognition methods under pose variations.
• Proposed an optimization technique using key pose samples within the recognition system's known faces.

Main Results:

• The new dataset effectively demonstrated the detrimental impact of head pose variability on existing face recognition algorithms.
• The proposed optimization method, incorporating key pose samples, significantly improved recognition system performance.
• Experiments confirmed enhanced effectiveness of optimized systems, specifically those using Multitask Cascaded Convolutional Neural Networks (MTCNNs) and ArcFace.

Conclusions:

• Optimizing known face sets with key pose samples is an effective strategy to mitigate performance degradation caused by head pose variability.
• The developed dataset and methodology provide valuable resources for future research in robust robot face recognition.
• This work advances the potential for reliable human-robot interaction in real-world applications through improved face recognition capabilities.