INTELLIGENT SYSTEM FOR TEACHING DRIVERS TO DRIVEA CAR USING TELEMETRY IN REAL TIME

Abstract

The article analyzes modern approaches to improving road safety, in particular the limitations of traditional monitoring solutions and driver assistance systems (ADAS), which mainly perform surveillance and warning functions. The feasibility of switching to intelligent driver training systems focused on the formation, correction and consolidation of safe vehicle control skills directly during movement is substantiated. The concept of an Intelligent Driver Training System (IDTS) is proposed, capable of operating in a mode close to real time. The key feature of the system is the integration of heterogeneous data about the driver, vehicle and road environment using hybrid telemetry, which combines the capabilities of a smartphone, OBD-II adapter and front video camera. The conceptual architecture of IDTS and its main functional modules are described: collection and time synchronization of data streams, pre-processing and multi-sensor fusion, interpretation of maneuvers and traffic events, as well as assessment of driving style. Driving style scoring is implemented based on a combination of expert logic rules (rule-based approach) and the Random Forest machine learning algorithm, which provides a balance between interpretability and adaptability of the model. Special attention is paid to the driver condition monitoring module, which uses EAR and PERCLOS indicators to detect fatigue and reduced attention, with further integration of these parameters into a comprehensive trip assessment. It is shown that the proposed architecture can be implemented on an affordable and mass hardware base without the need for specialized on-board systems. This makes IDTS a promising platform for use in driving school training programs, corporate systems for improving driving safety, and telematic services of insurance and transport companies. The main requirements and practical limitations of real-time operation are outlined, in particular regarding computing resources, power consumption, and acceptable delays, and the feasibility of distributing computations between edge devices (smartphone edge level) and server infrastructure for long-term analytics, personalization of learning, and accumulation of statistical models of driver behavior is substantiated.