Investigation of the Automated Object Recognition System for Visually Impaired People
Students Name: Pastukh Volodymyr Andriiovych
Qualification Level: magister
Speciality: Information Communication Networks Design and Administration
Institute: Institute of Telecommunications, Radioelectronics and Electronic Engineering
Mode of Study: full
Academic Year: 2022-2023 н.р.
Language of Defence: ukrainian
Abstract: The graduation thesis presents a prototype of an automated object recognition system for blind people. The object of study – facilities for people with visual impairments The field of research – health care technologies The purpose of the study – delegation of human functions to hardware and software tools In parallel with the development of technologies, the idea of transhumanism developed in scientific and engineering circles. The possibility of replacing the biological components of the body and human functions with artificial technical and software tools is becoming more and more popular in modern realities and opens up new perspectives. In this work, the history of the development of a separate branch in this field, health care technologies for people with visual impairments, was unfolded. The behavior and characteristics of a blind person, his life in society, and his needs were analyzed. Also taken into account is the biological component, the work of the relevant researched organs compared to a healthy one. This information was used to specify the system’s requirements and functions to meet the users’ specific needs. Variants of systems in a historical perspective, existing at this point in time, erroneous attempts to implement certain functions, postponed in the past due to technical limitations of the methods are analyzed. An ambiguous, but the key point in the perception of objects with which the system created for blind users should work was revealed, which are conventionally divided into explicit, directly processed in a holistic form, and implicit, used to increase accuracy, simplify processing, and scale functionality. Based on this, the development direction was determined and a new option was proposed, which consists in the use of modern data processing methods and their combination in one system with a specialized algorithm. At this stage of technology development, a large number of hardware and software options are provided for the implementation of such a system, which paradoxically does not simplify, but complicates it. A large number of attempts at different combinations are required, which requires specific resources and time, and makes it impossible to quickly develop a ready-made solution. This problem was solved at the stage of research and design of the system architecture. In the work, a study of the optimal solution was carried out, in the form of a large-scale software system, which includes easily replaceable components for processing various types of simple data, to perform all the main functions of the system. Each component can be isolated and used independently of the others, and works with primitive data and operations defined up to a certain limit, beyond which it is divided into smaller ones, in order to maintain homeostasis. Researched and created a basic unique algorithm for linking independent components, which works asynchronously with a dynamic number of data streams. A feature of the algorithm is a well-defined productive distribution of data to the corresponding operations of preliminary and final processing, without loss of integrity and delays in execution. The designed architecture based on the principle of connected modules, through the uniquely created algorithm of asynchronous data processing, has a list of clear advantages, which distinguishes it from other existing analogs, and prevents the problem of project development of this kind. This architecture allows you to combine and fully test various combinations of implementations of functions, and to improve the system based on intermediate results of self-generated data and feedback from user testing. The development of the system included the most modern approaches, which are not yet fully verifiable and determined for specific ways of use, but showed good results at the research stage. The use of machine learning algorithms and neural networks, to some extent, is a risky option, but acceptable in this case of the designed architecture. Computational power, uniqueness in approaches with sizeable raw data sets, and ease of implementation solidified the basis of further steps. The result of the work is a version of the object recognition system for people with visual impairments, in the form of a prototype for testing with all the requirements and functions determined at the analysis stage implemented. This initial version included the classification of visible objects in the external environment using a built-in neural network with direct output of the result to the user, description of the space using computer vision algorithms, determination of the depth of the image from the camera for further processing into a continuous picture and assistance in spatial orientation of the user. Also includes recognition of implicit macro objects of the environment in the form of global satellite navigation, location, and route determination. Provides an intuitive specialized interface for interacting with a blind user. In conclusion, an in-depth analysis of the research object and the system created based on its results make it possible to continuously develop the project until the limit of acceptable performance is found. The designed architecture and unique asynchronous processing algorithm can be adapted to other technical means in the future.