ANALYSIS OF APPROACHES TO DETERMINING THE SPATIAL POSITION OF OBJECTS USING COMPUTER VISION

Abstract

This paper analyzes and systematizes approaches to determining the spatial position of objects using computer vision, which is an important component of modern intelligent visual information processing systems. The relevance of this research stems from the widespread use of computer vision in fields such as the fashion industry, interior design, robotics, autonomous transportation systems, and augmented and virtual reality, where the precise localization of objects is critically important.

An analysis of approaches was conducted based on two classification criteria: the type of input data and the method of information processing. It is shown that monocular methods are characterized by simplicity of implementation and minimal hardware requirements, but are limited by the lack of direct information about scene depth. Stereoscopic approaches and methods using depth sensors provide higher accuracy due to additional spatial information, but require more complex hardware implementation.

Data processing approaches are considered: geometric methods based on mathematical camera models and spatial transformations; feature-based methods that utilize characteristic image elements; and deep learning neural network methods that ensure high performance under complex conditions and are capable of accounting for nonlinear data dependencies.

It has been established that hybrid approaches, which combine classical and neural network methods, allow for increased accuracy and robustness with acceptable computational complexity.

The study concludes that the most promising direction is the integration of different types of data and processing methods, as well as the development of hybrid models for operation under conditions of uncertainty and in real-time.