Stop Sign Detection

In September of 2017, the McMaster Engineering EcoCar 3 ADAS team was in the process of developing goals for it's final year in the EcoCar 3 competition. Having been involved with the team for 2 years, I continuted my work on the development of computer vision systems such as Stop Sign Detection.

Objectives

My objective was to develop a stop-sign detection algorithm with a higher success rate, and lower false-positive rate than previously developed systems.

Strategies

The stop sign-detection system was developed using MatLabs computer vision toolbox, and the 'Cascade Object Detector', a built-in function based on the Viola-Jones algorithm. To improve the detector results, I used an image training set with uniform resolution, and a greater number of images. Additionally, I pre-processed the training images to take advantage of features in the YUV colour system that make a stop-sign more prominent against the rest of the image.

Outcomes

After many attempts, a better detection system was achieved. By increasing the size of the training set, and by using images of uniform resolution, the detector produced fewer false-positives, and had increased successs in detecting stop-signs. See the video below for the final result. In the future, it may be beneifical to use an aggregated channel feature detection, to better extract and classify stop-sign features. This may result in a detection system with an even higher success rate, as more image features can be used to describe and classify objects.