Computer Vision enabled Adaptive Speed Limit Control for Vehicle Safety

Abstract

Over speeding, especially by heavy vehicles, is the primary cause of accidents in India. The reason can be attributed to the lack of a framework to maintain strict road safety rules. This leads to heavy vehicles occupying a high-speed lane, which leads to frustration and rapid lane switching among passenger vehicles. With the recent advancements in Computer Vision and IoT, it is possible to enforce such safety rules without on-ground personnel. In this paper, we have proposed an IoT-based solution for vehicle speed control that uses computer vision to detect the lane and dynamically limit the vehicle's speed, thus discouraging higher speeds of certain vehicles on specific lanes. We have manually labelled ~1.2 lakh images of the TuSimple lane dataset for training the models. We have provided CNN models as a baseline and used a pixel counting based SVM method for detecting lanes which achieved CNN levels of accuracy while being computationally efficient. Our proposed solution aims to automate the regulation of speed on a per vehicle basis, which can be very effective in reducing the number of accidents in India.