FotoNation’s Object Detection Engine (ODE) is the next generation of convolutional neural network based solutions for object detection, tracking and template matching that addresses current problems and limitations in the image processing pipeline.
ODE is used for computationally intensive tasks, with control and configuration performed in software and is upgradeable post silicon with network weights, delivering outstanding detection distance at a very low resolution and very high frame rate (60 FPS).
At the heart of this revolutionary IP core performance are two key segments: Deep learning imaging infrastructure with billions of ground truth data sets for training and validation and the testing framework with tens of millions of marked real world images for verification and testing.
Designed to work with minimum resources – very low processor and power usage – ODE accelerates and enables a wide range of features such as face or body analytics that can run in parallel.
10x speed acceleration compared with typical software implementations.
Decreases the CPU load (x100) allowing real time performance.
Fully reusable for other network architectures.
Extension for specific use cases and post-silicon quality improvement.
Real-time execution for detection or segmentation tasks.
Computer vision algorithms are at the foundation of some of the most outstanding implementations of today’s embedded systems.
Understanding images has become key to deploying or improving existing use-cases across various industries. The continuous technological improvements have led to highly specialized systems that can detect and track an object with incredible accuracy.
The Object Detection Engine is the first of its kind programmable engine that can be tailored to a specific scenario with virtually no limit of objects detected. It can adapt to any scenario or industry while delivering the best results, even under difficult illumination conditions or within systems with power consumption or bandwidth restrictions.
Fast and precise, with detection data available in every frame, our engine is the perfect choice to accelerate computer vision algorithms for face or body pose detection and tracking, face features detection, smile and blink detection, face recognition, and face beautification.
Besides these typical applications, the engine can be programmed to accelerate the detection and tracking of generic objects with no-trade-offs, leading to the same performance and precision on never seen before use-cases.