Traffic sign detection based on color and shape detection

Description

In this project, an algorithm for traffic sign detection based on color and shape detection is developed. The algorithm uses images taken by a low resolution camera mounted in front of a moving car. Two types of traffic signs, yellow warning signs and red stop signs, are tested and detection results are summarized. The conclusion is that color-based detection is sensitive to illumination condition and shape detection is sensitive to the complexity of the background.

The algorithm implemented in this project is based on color detection followed by shape detection using Hough transform. The algorithm takes an input image and pre-process the image through color enhancement. Then the algorithm perform color segmentation according to a target color. The output after this step is a binary image with regions containing the target color labeled. Then an edge detector is applied to each region. Hough transform is then performed on the edge image. Detection result is determined by looking at the distribution of peaks in Hough transform.

Several conclusions can be drawn by looking at the experimental results. First of all, illumination condition can vary from image to image and can greatly affect the results from both the color segmentation step and the edge extraction step. Low resolution image is another big challenge for effective edge extraction. And the complexity of background can result in great amount of outliers in Hough transform and therefore create huge error in shape detection.
Furthermore, currently the algorithm requires apriori knowledge of the color and shape of the traffic sign being detected. The algorithm will not be able to tell if the given information is incorrect.
Clearly, certain limitation presents in the current detection algorithm and detection results are highly dependent on the quality of the images. Machine-learning techniques can potentially improve the detection result and should be looked into for the improvement of detection accuracy.

MATLAB Implementation of segmentation-based disparity averaging

Efficient Graph-Based Image Segmentation