Real-time vehicular accident prevention system using deep learning architecture

Abstract

Approximately 20 million people die every year of road accidents, mainly caused due to ignorance of road safety norms and traffic rules. Drivers’ experience and prudence still have to be relied upon to prevent vehicle accidents. Here, an algorithm is developed to prevent T-bone accidents, rear-end and head-on collisions, pedestrians accidents. Using Raspberry Pi 4 Model B with 8 GB RAM as processing unit, importing OpenCV library into python 3; an intensive dynamic real-time algorithm for the effective rear end and adjacent collision avoidance module is developed on the framework of deep learning architecture(DLA) for image categorization. In this work, real-time object identification, distance estimation, and tracking of the instantaneous position are performed in all environmental conditions using a deep learning algorithm without any additional sensors. The Raspberry Pi NoIR Camera Module V2 keeps sending the captured frames in real-time to feed the MobileNet DLA for Single Shot Multibox Detection (SSD). The attributes of the preceding obstructions are utilized in such a way that the live-distance between vehicles and other objects is approximated successfully by providing the instantaneous positional information whereby generating warning signal on exigency with a processing speed of 0.09 s per frame (11 fps). The real-time output data are analyzed by multivariate analysis and a 3x3x9 repeated measures ANOVA. The result shows that Multimodal Early Accident Alarming System is highly effective for distractive drivers.