Dirt and Blockage Detection

Executive Summary

Cogent Dirt and Blockage Detection provides a set of computer vision algorithms which can be applied to highly distorted (fish-eye) cameras to detect a blocked camera image – a requirement of several carmakers.

Licensing Options

As with all of our software: Contact us to discuss how we can best support your project. Royalty-free licenses to our existing software may be available in certain arrangements.

Feature Description

Camera systems have become ubiquitous on new cars, and OEMs have come to expect more than just a camera. In many cases, advanced diagnostic capabilities are expect, including the ability of a camera (or an ECU downstream of the camera) to detect external blockage.

Our blockage detection algorithm is designed for real-world scenarios that could interfere with an automotive camera:

  • Opaque soil
  • Dust accumulation
  • Water droplets
  • Icing
  • Lens defocusing and damage
  • Fog
  • Tape, paste and plastic bags

The algorithm is optimized for embedded systems and includes two parallel pipelines: A fast (GPU/DSP) and slow (CPU) processing pipeline.

The algorithm is capable to detect dirt (a blockage) within about 10 seconds with less than 1% false positives in case of smooth gradients (sky, clouds, snow, sand) and less than 0.01% false positives in case of typical diverse environment.