SFPD: Simultaneous Face and Person Detection in Real-Time for Human-Robot Interaction

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Face and person detection are foundational in computer vision, essential for tasks like facial recognition and human action analysis.
• High detection rates and fast processing are critical for the overall performance of these recognition systems.
• Existing lightweight networks are often object-specific, necessitating a unified approach for simultaneous detection.

Purpose of the Study:

• To develop a unified framework for simultaneous face and person detection.
• To achieve state-of-the-art detection performance comparable to specialized networks.
• To maintain real-time processing speeds for both detection tasks concurrently.

Main Methods:

• Applied multi-task learning to combine face and person detection within a single network architecture.
• Developed a specialized training procedure and network design to overcome the lack of combined annotated datasets.
• Utilized an algorithmic approach to generate ground truths without manual annotation or compromising quality.

Main Results:

• The proposed Simultaneous Face and Person Detection (SFPD) method achieves a detection rate of 40 frames per second.
• Demonstrated a strong trade-off between detection accuracy and inference time.
• The framework successfully integrates both face and person detection capabilities.

Conclusions:

• SFPD offers a valuable and efficient real-time framework for computer vision.
• The method is particularly suitable for real-world applications, including human-robot interaction.
• The novel training and architectural approach addresses dataset limitations effectively.