This paper describes a computer vision system designed to perform an inventory of traffic signals. The system consists of five Ethernet synchronized cameras; the acquisition strategy allows us to take one image per camera every other second. We then use those five images to generate a panoramic image each second. Signal detection and recognition is carried out offline. Detection of traffic signal is done in the panoramic image using the Hough transform and enhancement of HSV color space. Traffic signal recognition is made by a combination of Haar wavelet and violates Jones classifier. Finally, we present experimental results using a database of one hundred images.