industrial applications Inspired by the NASA rovers that use stereo cameras to drive on Mars, we combined hardware, proprietary depth estimation algorithms, machine learning and power optimization expertise into a visual cortex (a "brain in a box") for machines that interact with the physical world.
  • The stereo camera captures slightly shifted left and right views that are analyzed by a perception stack.
  • The Krae-1 combines GPU hardware and FPGA based I/O interfaces to convert multiple video streams into actionable information at application-specific power budgets.
  • Simultaneous depth estimation and object detection is achieved using a combination of mathematics and deep learning based features.
Real-time visual commonsense is widely applicable to improve the capabilities, productivity and safety of industrial equipment.


Cajal Vision Engine

Krae Vision Engine The Cajal Vision Engine integrates algorithms, ML models, camera models and a well engineered API to provide complete software solutions for specific vision tasks. In the 1980s and 90s each computer game had its own custom rendering code for graphics. As games became sophisticated, the effort needed to build from scratch became prohibitive. Today, games are written using game engines such as Unreal or Unity. Assets are imported into game world or purchased from asset stores and the engine provides rendering, AI etc. By analogy, computer vision and ML are too complex to integrate into real-time applications. We provide a vision engine, pre-trained models and hardware and take care of details such as AMD vs Nvidia and ARM support.
  • Analyze scene, detect objects of interest
  • Extract actionable information
  • Pre-configured and optimized stacks of task specific ML models
  • Intrinsic parameters for wide variety of cameras
  • Software only solution for PC-like applications
  • Apps run on real-time vision API similar to games on a game engine API

Krae Vision Processor

Krae Vision Processor The Krae-1 is a ruggedized industrial computer optimized for our perception stack. This scalable multi-vendor GPU platform is available at power budgets from 1 to 100W with an easy to use API. We are open to customization based on your requirements and are currently using the following verticals to drive development:

Forklifts and industrial machines: Forklifts (also excavators, snow removers, sweepers etc.) are accident prone due to limited visibility and high center of gravity. Kraenion’s perception technology improves productivity and safety with faster docking at pallets, fewer tip overs on narrow ramps, and better situational awareness of people and objects around the vehicle.

Wheelchairs and mobility scooters: Most wheelchair users experience at least one accident (tip-and-fall, accidental contact, etc.) because of difficulty in maneuvering on driveways, uneven surfaces, and in tight spaces. Our technology can enhance wheelchairs with smart vision to improve safety, and enable better mobility for an aging baby-boomer population.