ADAPTIVE VISION AI

Abstract. As of today, computer vision systems are continuously developing and systematically improving. Machines see visual content in the form of numbers, in which each pixel represents its own piece of information. Computer vision, as a component of artificial intelligence, allows machines to see, observe and understand everything. It enables computer systems to obtain useful information from digital images, video, visual data and perform programmed actions. Computer vision technologies rely on pattern recognition, machine learning, and neural networks to allow computers to break down images, interpret data, and identify features. Tracking moving objects and their identification is a difficult task, as it requires the accuracy of pattern recognition. An untrained computer vision algorithm is unable to understand the relationship between the shapes in the image and the objects. Therefore, the algorithm must be trained. The paper considers models that are trained on a high-performance computing cluster with GPU support. The developed open source software allows detection, tracking and recognition of blurry moving objects with the help of artificial intelligence that adapts to any video camera. A significant increase in accuracy is achieved thanks to machine learning.