This set of technologies is grouped into two parts:
1. A powerful deep learning algorithms package developed through ArcSoft’s years of experience and research. This package can achieve high-speed, accurate human detection, tracking and recognition.
2. ArcSoft’s visual algorithms package, which can evaluate characteristics of the human body with high-precision.
Human Body Detection: Versatility is a key necessity for this feature – It needs to function in various kinds of lighting conditions, especially dim-lit (less than 4 LUX) environments, side-lighting, backlighting, and raised or lowered lighting sources. It must also be able to operate at all angles of view without blind spots and support detection even when the subject is obstructed from view and reappears. Finally, optimal technologies should also support quick movement and wide-range zooming, and be able to track multiple subjects in realtime. ArcSoft has used many years of research and development to produce technologies which can accomplish each of these goals, while being extensively optimized for various hardware devices to achieve high-performance operation with low power consumption.
Human Body Analysis: Various solutions are provided for single depth camera setups and multi-camera setups. In order to assure a comfortable experience for the subject of analysis, we have extensively optimized performance so that results can be processed in under 10 seconds for multi-camera configurations, and in under 50 seconds for single-camera applications.
Posture Recognition: Deep-learning algorithms and high-speed front-end classifiers have been combined to achieve high-efficiency in quickly detecting common postures.