Undergrad Research Project - Traffic sign detection in night images

Fall 2017

Adolfo Karim Victoria Higueros
John Dolan
Project description

The objective of this project is to detect common traffic signs in road images and estimate their retro-reflectivity. The Navlab group has a large dataset of images taken from within a vehicle looking out the front window. Many of these were taken at night. Adolfo Victoria will learn to use a deep learning based object detector (e.g. faster-RCNN) to detect the traffic signs. He will read publications and check open source software repositories for a suitable program. He will train it for this particular class of objects. Next the student will develop methods to estimate the brightness of the sign in the image and check if the sign is within the light cone of the headlights. Together with the approximate distance between sign and vehicle one can then estimate the retro-reflectivity. The student will evaluate his methods by comparing them to ground-truth data. Finally, the code, documentation and usage instructions of his developments need to be committed to a Github repository.

Return to project list