Designing an Embedded Tra c Dynamics Meter: A case study

Abstract

Costa Rica is actually facing a severe problem with people transportation caused by tra c congestions, which degrade people life quality due to the time invested during daily commutation, and the stress during the driving. One of the reasons is the incapability of automatic tra c signals to respond to the change in tra c dynamics during the day. In order to reduce the impact of automatic signalling on tra c congestions, this work proposes to quantify tra c dynamics using average speed, tra c ow, tra c density, and sensor occupancy using Computer Vision. However, there are constraints such as Internet connection unavailability to upload data to the Cloud, and absence of enough energy to power a high-performance computer up in some places. Therefore, the challenge of developing this meter implies to implement the system on a low-power platform. The application is optimised using Approximate Computing paradigm to tackle this problem, after noticing that it is possible to enhance runtime and power consumption sacri cing results accuracy. After applying approximations, it is possible to improve the application runtime up to 2:55x introducing an error of 11:5% on output results.