Dirt and Blockage Detection
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.
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.
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
- Lens defocusing and damage
- 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.