Enhanced Real-Time Computer Vision and Intelligent Decision-Making for Autonomous Vehicle Applications

Abstract The trends of autonomous driving and intelligent transportation systems are increasing exponentially in the global context and sooner or later, it will overtake the existing transport sector. Artificial Intelligence (AI) plays a paramount role in its strengthening and deployment. Computer vision and decision-making are some of the primitive tasks executed by AI techniques. The prominent challenge arises when the detection contains false alarms which leads to inaccurate computer vision and correspondingly, wrong and imprecise decision-making. This paper presents an enhanced computer vision and decision-making framework for real-time intelligent transportation system applications and assisting the scenario of autonomous driving with minute false alarms. Existing technologies are unable to detect animals in real-time which is a major fallback for autonomous vehicle applications. The system involves a developed computer vision technique based on deep learning YOLO v6 and optimized with stochastic variance reduced gradient approach capable of detecting bike, car, mini truck, cow, dog and pedestrians with minimal false alarms. The developed technique was tested on real-world road networks using 50 set of vehicles. The mean average precision (mAP) scores for all classes accounts to 0.9783 with a frame rate of 87 frames per second deployed on NVIDIA GPU. The developed system showed precise detections in both day and night time and presents a strong baseline for advanced autonomous vehicle applications for real-world transportation scenarios.