Auto-Alert: A Spatial and Temporal Architecture for Driving Assistance in Road Traffic Environments

Over the last decade, the Advanced Driver Assistance System (ADAS) concept has evolved prominently. ADAS involves several advanced approaches such as automotive electronics, vehicular communication, RADAR, LIDAR, computer vision, and its associated aspects such as machine learning and deep learning. Of these, computer vision and machine learning-based solutions have mainly been effective that have allowed real-time vehicle control, driver-aided systems, etc. However, most of the existing works deal with ADAS deployment and autonomous driving functionality in countries with well-disciplined lane traffic. These solutions and frameworks do not work in countries and cities with less-disciplined/ chaotic traffic. Hence, critical ADAS functionalities and even L2/ L3 autonomy levels in driving remain a major open challenge. In this regard, this work proposes a novel framework called Auto-Alert. Auto-Alert performs a two-stage spatial and temporal analysis based on external traffic environment and tri-axial sensor system for safe driving assistance. This work investigates time-series analysis with deep learning models for driving events prediction and assistance. Further, as a basic premise, various essential design considerations towards the ADAS are discussed. Significantly, the Convolutional Neural Network (CNN) and Long-Short-Term-Memory (LSTM) models are applied in the proposed Auto-Alert. It is shown that the LSTM outperforms the CNN with 99% for the considered window length. Importantly, this also involves developing and demonstrating an efficient traffic monitoring and density estimation system. Further, this work provides the benchmark results for Indian Driving Dataset (IDD), specifically for the object detection task. The findings of this proposed work demonstrate the significance of using CNN and LSTM networks to assist the driver in the holistic traffic environment.