REAL-TIME SMART VEHICULAR SYSTEM USING GEOMATICS AND IoT TECHNIQUES FOR CONNECTED VEHICLES

Abstract

With increasing population, transportation sector experiences heavy influx of density creating immense challenges in its management. Range of technologies are advancing exponentially at a rapid pace resulting into the evolution of advanced driver assistance systems (ADAS) laying a strong platform towards the intelligent transportation systems (ITS). This rapid growth in the technology varies from context to context giving rise to uncertain scenarios and circumstances in the transportation sector. Dense population results into higher number of vehicles on road making undesirable situations and hence, affecting the efficiency of automated features. With developed countries adapting self-driving cars and automated vehicles, India lacks behind due to the presence of erratic contextual characteristics like irregular pedestrian movements, ill traffic management, high density of vehicular movements, sudden and unexpected animal movement as well as presence of potholes on the roads. This study focuses on countering such unwanted characteristics especially on the Indian transportation context. Furthermore, the study aims at assisting automated and self-driving cars to arrive sooner or later in India and intends to deliver enhanced ADAS features namely real-time computer vision, advanced transportation safety and real-time decision-making capabilities. The Indian automobile market depends upon low-end vehicles as majority of the population utilize vehicles below 1 million rupees. This results into the fact that high-end ADAS-rich vehicles like Tesla and Volvo are of limited concern amongst the Indian automobile market. It is therefore essential that a downscaled system is in place for coping up with this challenge and providing features rich in ADAS. Downscaled hardware along with optimized sensors capable of performing ADAS tasks in real-time are the factors that allow the system to pave the way for self-driving cars to arrive effectively. Long-throw camera sensor with resolving entities at a distance and detecting them efficiently is the foremost skill required. Furthermore, executing the entire workflow in the GPU environment enables the system to perform the task of detection quickly and smoothly. Artificial Intelligence is a spotlighted field of research that is evolving at an exponential rate covering a wide spectrum of applications and the domain of transportation is a no exception. Computer vision is an indispensable step in the fields of ADAS and ITS as it acts as “the eyes” for the self-driving and autonomous vehicles. This creates a dominant necessity that this step provides efficient and precise outputs. Computer vision in the context of transportation refers to the detection of entities present in the traffic scene namely cars, bikes, mini trucks, pedestrians, cows and dogs. Majority of the studies have focused on vehicles and pedestrians but the classes of cows and dogs viz. stray animals are extremely essential in the Indian context as they are present in abundance in the transport network. Real-time computer vision in this study was implemented based on the YOLO v4 (You Only Look Once) architecture of deep learning based object detection framework. The developed algorithm detected the aforementioned classes with high efficiency and precision scores. With dense and irregular traffic patterns, it is primitive to achieve safety in the transport network. Along with the accidents due to vehicles, pedestrians and animals, potholes are a major contributor for creating undesirable accident-prone scenarios. Mobile App was designed capable of fetching the location of the pothole and relaying it to the central cloud for storage of the pothole’s geo-coordinates. Integration of geomatic techniques assists in creation of geospatial database as well as applying the techniques of hotspot mapping as well as spatial decision support systems. For this, the vehicle’s GPS was consistently connected to the cloud-server through IoT connectivity and the location was monitored at consistent epochs. Alarm was triggered whenever the vehicle approached the hotspot zones. The hotspot zones comprised of pothole hotspots as well as animal hotspots because these two are the major contributors in origination of undesirable scenarios leading into the circumstances of fatal accidents and loss to the humans, animals as well as the infrastructure. The overall combination of real-time computer vision, GIS-based location mapping, hotspot creation and integration of IoT-cloud approach provided efficient and effective assistance to the on-board drivers. Additionally, the system delivered precise alarms and alerts for avoiding pothole and animal-led accidents ensuring transportation safety. With rapid advancements in the field of ADAS features, internet connectivity is highly emphasized and with consistent IoT link, the entire system executed data push-data pull with the server stored on the cloud. The cloud-server is stored on the domain https://www.theintelligentcar.com which contains the server link, geospatial database as well as the location hotspots. The study successfully assists the entire Indian transportation network ensuring real-time computer vision, transportation safety, enhanced ADAS assistance functionalities as well as creating a strong and robust platform for advanced ITS applications like self-driving and autonomous vehicles.