A Finite State Machine-Based Multisensor Selection Architecture for Real-Time Advanced Driver Assistance System for Indian Road Conditions
Abstract
Advanced driver assistance systems (ADASs) trigger warning signs for potential safety violations by using multiple sensors and warning classes, each requiring complex computer vision and signal processing pipelines. This article has twin objectives of improving computational efficiency by processing selected sensors for ADAS, and generalizing the ADAS specifications for Indian traffic conditions. First, this article presents a finite state machine (FSM)–driven multisensor selection architecture that activates and deactivates sensors and warning classes as the vehicle moves. The proposed system integrates five camera feeds and dual radar sensors to enable activation of warning classes. The FSM governs system transitions based on vehicle speed and driving conditions, dynamically managing sensor activation. The system deactivates sensors and warning classes that are not possible given the driving state. Second, this article generalizes the definition of ADAS warning classes to model Indian traffic conditions. We separately account for different vehicle types, stressing that motorbikes are the main entity in Indian traffic. We also account for large lateral movements in Indian traffic, attributed to loose lane discipline, and include lateral intent in determining the ADAS warning. The architecture was implemented on an NVIDIA Jetson AGX Orin (64 GB) operating in 30 W power mode and evaluated over a 100 km highway drive (out of 572 km of data collected), with controlled, staged validation scenarios. Comparative benchmarking against continuous parallel execution activating all sensors and modules shows that the FSM-based activation reduces CPU and GPU utilization by up to 18.7%, while sustaining stable real-time performance at 25 FPS. The results validate the proposed framework as a computationally efficient and deployment-ready ADAS architecture for resource-constrained embedded automotive platforms.