Embedded System for Detection, Recognition and Classification of Traffic Signs

Abstract

This study concerns the development of an embedded system with low computational resources and low power consumption. It uses the NXP LPC2106 with ARM7 processor architecture, for acquiring, processing and classifying images. This embedded system is design to detect and recognize traffic signs. Taking into account the processor capabilities and the desired features for the embedded system, a set of algorithms was developed that require low computational resources and memory. These features were accomplished using a modified Freeman Method in conjunction with a new algorithm "ear pull" proposed in this work. Each of these algorithms was tested with static images, using code developed for MATLAB and for the CMUcam3. The road environment was simulated and experimental tests were performed to measure traffic signs recognition rate on real environment. The technical limitations imposed by the embedded system led to an increased complexity of the project, however the final results provide a recognition rate of 77% on road tests. Thus, the embedded system features overcome the initial expectations and highlight the potentialities of both algorithms that were developed.