AEROS

Automatic Recognition of Traffic Scene Objects

Inspection vehicle with several cameras.

Inspection vehicle with several cameras.

Project Description

Our goal was to automatically recognize and locate important traffic scene objects like traffic signs, lane markings and protection planks. Such information is very valuable to ensure driver safety. It is also helpful for the optimization of traffic routes.

Our approach allows for frequent inspections of traffic scenes. For example, we can determine if all traffic signs are still located at their correct positions or if they have been damaged and became unreadable due to environmental influences. We analyze images from street inspection cars, which are equipped with multiple cameras. The detected traffic signs are located based on GPS information and visual distance measurements.

Traffic Sign Recognition

Various algorithms are used to evaluate color and shape information for the recognition of the traffic signs. Our system recognizes 43 different German traffic signs from multiple views. It is also able to discriminate between visually very similar signs.

Advantage of the acquisition and evaluation of image sequences with AEROS: High detection rate of traffic signs.
© Photo ITWM

Advantage of the acquisition and evaluation of image sequences with AEROS: High detection rate of traffic signs.

Our department worked closely together with our project partners from the industry and the University of Applied Sciences in Kaiserslautern. We evaluated state-of-the-art algorithms from the field of visual object detection and modified them so that specific project requirements were met. Within the scope of this project we supervised multiple bachelor theses and shared and discussed our results on a workshop. This work is the foundation for our prototype, which can reliably recognize 43 different German traffic signs. During the development of our software we focused on modularity, scalability and interoperability.

Traffic Sign Recognition

Architecture

The architecture of AEROS is based on a core component, which is responsible for the execution of different detection algorithms with pre-defined input and output interfaces.
The modules are arranged in a tree like structure and executed in parallel. The detection results are then combined using voting. Each detector looks for possible regions of interest in the image and produces scores that reflect probabilities on how likely an object is contained in a specific region. The voting then decides if a candidate region is accepted as a valid traffic sign.

Additionally, we can also use prior knowledge to rule out conflicting traffic sign setups such as if a stop sign is found above a right-of-way sign.

Architecture of AEROS
© Photo ITWM

Architecture of AEROS

Sponsorship

Sponsored by Federal Ministry of Education and Research - BMBF
2010-2013