Algorithm for Analyzing Noisy Video Data for Autonomous Robots Navigation

Autonomous robots are widely used for various purposes, including monitoring, research, defense, emergency situations, etc. The main advantages of autonomous robots are mobility, ease of deployment and real efficiency. The problem of autonomous robot navigation based on data obtained from a video channel is relevant. The study aims to implement an algorithm for analyzing noisy video data to detect key points of navigation objects. Traditional approaches to detecting key points of navigation objects apply iterative algorithms without the use of neural networks. YOLOv11 neural network model and RGB images as input data can solve the problem. The work results in finding the coordinates for the key points of navigation objects. The algorithm test data on the dataset and data obtained from testing the demonstrator of an engine with a central body are provided. The OKS (Object Keypoint Similarity) metric is used. The study results in developing a neural network, trained to detect navigation objects and their key points. Defining the key points of objects enables to correct inertial navigation systems. The obtained neural network is resistant to various noises, including overlaps.