Dual-V-Sense-Net (DVN): Multisensor Recommendation Engine for Distraction Analysis and Chaotic Driving Conditions

Abstract

The Advanced Driver Assistance System (ADAS) features provide better safe driving dynamics reducing road accidents and collisions. Typically, these include lane assist, automatic night vision, adaptive cruise control, driver risk analysis primarily seen in high-end vehicles. Furthermore, such a proactive mechanism alerts the drivers in unsafe/ dangerous situations and enable drivers to respond with immediate actions. However, such ADAS-based solutions do not work very well in heterogeneous, less-disciplined, and chaotic road traffic environments, as seen in Asian countries such as India, Bangladesh, Burma, Nepal, etc. In this regard, this paper aims to propose an advanced integrated deep learning framework <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dual-V-sense-Net (DVN)</i> to provide recommendations for safe driving. A new dataset - Behavioral Understanding of Driver’s-Distraction Dataset for India (BUDDI), is proposed for the behavioral understanding of driver’s distraction. This proposed recommendation framework is designed in a two-step manner. The first step involves an in-vehicle driver distraction analysis with real-time frontal and cross-view naturalistic driver data for driver distraction analysis. Second step is to perform an exhaustive experimental evaluation is carried out for the tasks such as road traffic scene perception, road navigable area prediction with vehicular temporal dynamics, lane assist, safe drivable area navigation and driver distraction alert. A decision block modeling helps in safe/unsafe driving predictions considering the feature importance. We present comprehensive experimental results for road segmentation, vehicular road lane categorization, obstacle recognition, and driver behavioral pattern prediction using the BUDDI and Indian Driving Dataset (IDD) datasets.