Computer Vision News - May 2016

Project COMPUTER VISION NEWS Solution It is thus necessary to employ more advanced techniques, coming from the field of machine learning. Several methods are available to do this and useful databases can be easily found on the Internet. Among the learning methods which deserve to be mentioned: deep vision and the Viola & Jones object detection framework. The above procedure have only detected the round sign within the image, without addressing yet the task of identifying the number, which is more subtle. At this stage we can continue the work in two different ways: one way would detect the circle itself and locate the area where the digits are situated, identifying the number itself with relative ease thanks to the finite scope of existing speed limits, generally ranging from 30 to 130 km/h with intervals of 10; the other way would require a binarization of the digits found within the sign, after which they are separated and moved to an OCR procedure which will identify them with the highest precision. A short demo of this application can be found in the video below. The use of machine vision offers a great flexibility across systems. Developing computer vision software usually starts with a research stage, where algorithms are developed using Matlab or Python. The learning models are being trained and the trained models can subsequently be used in real-time. Next, the algorithm is usually implemented in C++ by using open source libraries like OpenCV. The same C++ code can be used in multiple platforms like Windows or Linux and even by Android or IOS applications, since mobile platforms always support running efficiently C++ native code parts. 11