Body detection and segmentation is the natural extension of face centric computer vision technologies.
This technology is the foundational to the next generation of imaging products and the key to environmentally aware intelligent cameras that are versatile, easily customizable and resource efficient.
Our hybrid software/hardware implementation assures the best mix of performance, power consumption and flexibility while being a scalable, cost-effective and future-proof solution.
Our technology will not only be at the heart of advanced tracking systems for action cameras and drones, but it will also be able to play a key role for body based imaging techniques such as single camera still and video bokeh, background replacement and portrait enhancement. Imaging analytics scenarios can also be tackled for event detection, congestion analysis or AR/VR enhancement of user experience.
Real time execution via hardware accelerated convolutional neural networks (CNN) inference for multiple active and static poses as well as in situations where significant occlusions are present.
Next generation segmentation.
Robust and performance focused solution.
FotoNation body analytics solution can deliver 30 FPS in real-time with an extremly low power consumption when a dedicated IP core is used.
The video segmentation runs real-time at 30 FPS in preview mode on Android ARM v8 cores while the still segmentation runs under one second on Android ARM v8 cores.
The performance tests scored better than 80% detection rate for people on Pascal VOC and better than 90% recall rate for body segmentation on Pascal VOC.
The solution can detect a person standing, walking or running facing any direction: front, back or side with a negligible CPU load (only in the case when the detection effort is done entirely on the IP core).