DYNAMIC CURVE WARNING SYSTEM FOR HETEROGENEOUS MIXED TRAFFIC CONDITION

Abstract

Sharp curves are hazardous because of narrow, poorly maintained, and limited visibility roads. Drivers in heterogeneous traffic may engage in dangerous driving behaviors such as overtaking on sharp curves or driving at high speeds. As per the Ministry of Road Transport and Highway (MoRTH 2016), 21% of traffic accidents occur due to the overtaking of vehicles. The main reason for accidents on curve roads is due to the high speed of the driver or unawareness of the oncoming vehicle to the curve. Lack of communication between the ends of the curve and visibility problems around the curve make hazardous situations. Existing methodologies like convex mirrors, sensor-based collision avoidance systems, headlights, and horns are used to reduce collisions. In modern transportation systems, in-vehicle and in-road technology are important for collision avoidance. User perception studied through a questionnaire survey is first conducted to analyze how the users perceive the new concept of the driving assistance system instead of the conventional system (eg: convex mirror) that are commonly provided. For the detection process, vehicle detection algorithms (YOLOv5l) are used. Vehicle detection involves analyzing an image or video to identify the presence of vehicles by detecting patterns, shapes, and colors that correspond to vehicles, and then distinguishing them from the background. The annotation and augmentation process helps to improve the accuracy of the model. Curve collision warning systems use cameras and signal boards, to detect when a vehicle is approaching a curve too quickly, and then issue a visual warning to the driver to slow down or take other corrective action to avoid a potential collision. The accuracy and precision obtained for the curve collision warning are analyzed and both results are above the acceptable limit. So, this collision avoidance system is essential in preventing accidents and saving lives by detecting and responding to the potential collision in real-time.