The CogniSight chipset is the combination of a configurable image recognition engine on FPGA and neurons ready to learn and recognize the vectors generated by the engine. Depending on the application, the engine can extract signature vectors directly from the video and submit them to the neurons, or a memory chip might be used as a buffer to extract complex or multiple signatures per frame before submitting vectors to the neurons.

View introduction as a pdf, or as a slideshow.

Default Specifications

• CogniMem neural network chip with 1024 neurons
• Actel IGLOO FPGA with 600K gates programmed with the default CogniSight engine including:
  • CogniMem interface controller (neurons, digital input bus and reco-logic)
  • I2C master controller to interface with a sensor or actuator
  • USB and RS485 controller
  • Acquire video frame to memory
  • Transfer memory frame to host
  • Load knowledge from memory

Options

• Feature extraction
• Find objects
• Generate transform image

Package

• CS-CM1K bundle
• CS-CM1K-SDRAM bundle
• CM1K

The CogniSight chipset can address simple applications with very practical hardware implementations

• Trainable vision sensors

• Low-cost, low-power and small foot-print

• Embedded logic for image learning and decision making

The CogniSight chipset can address applications re-known as complex with a re-known non-linear classifier and a parallel neural network expandable at will

• Cope with fuzzy and ill-defined problems

• Adapt to contextual variations

• Multiple sensor and/or multiple expert systems

• Hypothesis generation and decision making process