As intelligent cameras and computer vision technologies continue to overtake conventional imaging in every day products, the difficulty of developing and optimizing new vision systems is becoming painfully clear. Simply tuning the ISP for image quality requires deep expertise and can be onerous and time-consuming at the best of times, and the task is further complicated when incorporating innovative optics, sensors and computer vision (CV) tasks for every product. The time has come for a new approach.

Introducing CRISP-ML

CRISP-ML uses machine learning methods (e.g. Deep Learning) to automatically optimize your full imaging and vision system. CRISP-ML can effectively combine large real-world CV training data sets with standards-based metrics and chart-driven KPIs to holistically improve the performance of your vision systems; this can be done across combinations of components and operating conditions previously deemed as unfeasible. By exploiting our innovative machine-based approach, CRISP-ML automates the tuning steps that are otherwise painful, costly, and time-consuming.

CRISP

 

systemoutput

Optimizing component selection, image quality, and computer vision tasks is a significant challenge.

In addition to the many parameters of the ISP, a classical vision algorithm typically has many parameters such as thresholds, weights, and filters that need to be tuned for the best performance on the CV task. Machine learning based vision algorithms use data sets together with optimization to “learn” (or “tune”) many parameters automatically. CRISP-ML extends this paradigm to “learn”/”tune” the machine learning weights together with ISP parameters in the same process.

Whether you are building vision systems that handle event detection, process control, autonomous navigation or automatic inspection, CRISP-ML will provide the ISP and CV optimization required to ensure the corresponding applications will succeed in the field.

Easily adapts to new optics, sensors, CV tasks and applications.

Optics

  • Stereo cameras
  • Array cameras
  • Lenslet arrays
  • Light-field cameras
  • Optical masks
  • Coded-aperture
  • Etc.

Sensors

  • RGB-IR
  • Foveon X3
  • 3CCD
  • New Bayer filters
  • Multispectral imaging
  • Medical scoping
  • Etc.

         CV Tasks

  • Event detection
  • Process control
  • Navigation
  • Object modelling
  • Automatic inspection
  • Etc.

Applications

  • ADAS
  • Security
  • AR/VR
  • IoT
  • Wearables
  • Robotics
  • Mobile
  • Etc.