Smartphones como plataforma de inteligência artificial de borda para assistência ao condutor: análise de desempenho computacional, térmico e energético

Abstract

Advanced Driver Assistance Systems (ADAS) have gained increasing relevance in the automotive industry, aiming to improve driving safety and comfort. At the same time, advances in Artificial Intelligence (AI) and the evolution of smartphones have opened new opportunities to implement these technologies directly on mobile devices, promoting an accessible and economically viable solution. This dissertation analyzes the computational, thermal, and energy performance of deep learning algorithms applied to smartphones for ADAS, taking into account resource limitations and real-time processing requirements. In addition to a systematic literature review, an experimental analysis was conducted using standardized benchmarks (AI Benchmark and Burnout Benchmark) and an ADAS application developed specifically for this study. The experiments covered devices from different categories (premium, mid-range, and entry-level), highlighting significant differences in sustained performance, energy efficiency, and real-time inference capability. The results demonstrated that smartphones equipped with dedicated hardware accelerators, such as GPUs and NPUs, exhibit superior performance and greater efficiency in realistic usage scenarios. Finally, practical recommendations were presented for technical optimization of the models to improve their viability in real applications, directly contributing to advances in vehicle safety and mobile computing.