Road and Object-Detection based Assistant for Driving (R.O.A.D)

Abstract

Humans while driving have a disadvantage of not always being attentive (whether it be changing the radio or of being tired etc.) while a computer, if trained, can always be fully attentive at detecting lanes. Driver support system is one of the most important feature of modern vehicles. This is to ensure driver safety and decrease the vehicle accident on roads. The most important aspect of such a system is its computer vision-based software. This software detects and recognizes the lane in which the car is currently in and also detects objects in the observable-environment of the car/software. But such a system is not suited for Pakistani roads due to the traffic norms and low maintenance of road conditions, so a system is required which can perform well under Pakistani road conditions. Hence R.O.A.D is developed which is a desktop-based python application, which detects lane and objects from a dash cam video using CNN based trained models. User uploads a video which is broken into frames and pre-processing is applied to each one. Then the model detects the lane from the frame and creates a visualization at the detected pixels. Then the object detection model detects the objects, and the frames are compiled back into a video format and the detected video is displayed to the user.