An AI-driven object segmentation and speed control scheme for autonomous moving platforms

Abstract

In recent times, Autonomous Moving Platforms (AMP) have been a vital component for various industrial sectors across the globe as they include a diverse set of aerial, marine, and land-based vehicles. The emergence and the rise of AMP necessitate a precise object-level understanding of the environment, which directly impacts the functioning like decision making, speed control, and direction of the autonomous driving vehicles. Obstacle detection and object classification are the key issues in the AMP. The autonomous vehicle is designed to move in the city roads and it should be bolstered with high-quality object detection/segmentation mechanisms since inaccurate movements and speed limits can prove to be fatal. Motivated from the aforementioned discussion, in this paper, we present inspect (velocity-inspect), an AI-based 5G enabled object segmentation and speed limit identification scheme for self-driving cars on the city roads. In inspect, the Convolutional Neural Network (CNN) based semantic image segmentation is carried out to segment the objects as interpreted from the Cityscapes dataset. Then, object clustering is done using the K-Means approach based on the number of unique objects. The semantic segmentation is done over 12 classes and the model outshines concerning state-of-the-art approaches for various parameters like latency, high accuracy of 82.2%, and others. Further, K-Means clustering based Speed Range Analyser (SRA) is proposed to determine the acceptable and safe speed range for the vehicle, which is computed based on the object density of every object in the environment. The results show that the proposed scheme outperforms compared to traditional schemes in terms of latency and accuracy.