Author: LUKA BACEVINA
Company/Institution: University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture (FAMENA)
Country: Croatia
e-mail: lukabacevina168@gmail.com; lb219467@stud.fsb.hr
web: www.fsb.unizg.hr
Mentor: Prof. dr. sc. ZELJKO SITUM
e-mail: zeljko.situm@fsb.unizg.hr
A computer vision system was developed to recognize light-signal signs on traffic lights for pedestrians for the purpose of helping blind people when crossing the road. A portable system was developed in the form of an Android application, which consists of a camera and an integrated computer system that processes data from the camera in real time. Such a system can help blind people to navigate more easily and independently in urban environments. The developed system is based on convolutional neural network technology and can be easily learned and modified to recognize and track other objects using machine vision. The application contains an image preprocessing program using the sliding window method and speeding up processing using a graphics processor. The developed application is functional in real situations, easy to use and robust enough for the basic purpose. Computer vision systems, with the help of advanced algorithms and machine learning, have the ability to recognize objects and actions in real time and can be used in various fields, from medical robotics to autonomous vehicles.