Automotive Machine Perception

Lorem ipsum dolor sit amet, consectetuer adipiscing elit, sed diam nonummy nibh euismod tincidunt ut laoreet dolore magna aliquam erat volutpat. Lorem ipsum dolor sit amet, consectetuer adipiscing elit, sed diam nonummy nibh euismod tincidunt ut laoreet dolore magna aliquam erat volutpat.

Initial Situation

 

Goal of the project was to implement an Object classification and localization software module in traffic for Cars, Trucks, Motorcycles Bicycles and Pedestrians using a Mono Camera mounted in the windshield.

Requirements:

  • Runtime >10 FPS
  • Accuracy >95% TP in ego lane
  • Object: 2-wheelers, Person, Truck, Car
Figure 1: Forklift with our stereo smart sensor for localisation and mapping

Solution

 

  • Development of Deep Autoencoder for Image compression
  • Evaluation of different image compression algorithms for faster loading/storing times

 

 

 

Results

 

  • Faster loading times (x10) by compressing the reference images with the Deep Autoencoder
  • Similar accuracy for the rotation estimation
  • Concept for improved rotation matching strategy