Real-Time Driving Assistant using Deep Learning & Object Tracking

Abstract

In this final year design project, we present a real-time driving assistant system that leverages deep learning and object tracking algorithms to provide assistance to the driver of the host vehicle and make our roads safer. The system utilizes a customized YOLOv8 model for realtime detection of objects in a live video stream. Object tracking is achieved through ByteTrack algorithm over consecutive frames, enabling the system to obtain tracking IDs and bounding box data for each detected object. The information is used to track the position of objects over time and derive a closeness factor, which quantifies the rate of change in proximity of the object to the host vehicle. By using the object’s location in the frame along with closeness factor, the system can assess the rate of approach of detected objects. Using this information, the system notifies and warns the driver of potential hazards in realtime. Additionally, if the driver fails to react to the warnings, the system is equipped to apply brakes automatically, which can mitigate the risk of collisions. Our proposed system addresses the unique challenges posed by Pakistani roads, where traditional driving assistant systems often struggle to perform optimally. The real-time capabilities of our system provides drivers with timely and accurate alerts, significantly enhancing road safety in challenging traffic conditions.