Mrak V. Methods and means of recognition of moving objects in telecommunication systems in real time

The dissertation is dedicated to the development of methods and tools for real-time recognition of moving objects in telecommunication systems. In the first chapter, «Problematic Issues and Analysis of Technologies for Building Video Surveillance Systems» the features of both conventional video surveillance systems and systems incorporating video analytics functions are described. A review analysis of object detection algorithms and object tracking algorithms is conducted, leading to the proposal of a classification method for object tracking algorithms. The second chapter, «Development of methods and models for face recognition and traffic compression using a background subtraction algorithm for stationary video surveillance systems," presents an efficient background subtraction algorithm based on ViBE for complex backgrounds. This algorithm integrates an improved ViBE, which employs cumulative averaging and a pixel counting mechanism for rapid ghost removal, as well as an adaptive ViBE, which calculates an adaptive radius depending on background changes. Additionally, a video compression algorithm is proposed for footage captured by resource-constrained cameras. It utilizes motion analysis to eliminate unnecessary pixels and image frames. A motion-based face recognition method is also introduced, focusing not only on face detection and recognition stages but also on detecting moving objects, thereby reducing the amount of data requiring processing. A software model for data processing in surveillance systems is developed, and the efficiency of Spark technology is demonstrated. Spark ensures fast algorithm execution even with increasing data volumes due to computational optimizations. In the third chapter, «Investigation of developed methods and models for face recognition and traffic compression using a background subtraction algorithm» an in-depth study of the developed ViBE-based background subtraction algorithm for complex backgrounds is conducted. The proposed method for compressing video captured by resource-constrained cameras is also examined. By analyzing video frames pixel by pixel, this method achieves an average test data compression of 30%, while preserving all information crucial for analyzing the behavior of moving objects. This significant file size reduction enhances monitoring capabilities by minimizing bandwidth requirements and facilitating efficient data transmission from resource-limited devices. Deploying this method in cameras has the potential to significantly advance monitoring and storage optimization by enhancing the efficiency of existing systems. Furthermore, the proposed motion-based face recognition method is analyzed. This approach offers several advantages, including data volume reduction, increased speed, reduced energy consumption, adaptability, and improved performance. Motion detection allows processing only those parts of an image where changes occur, significantly reducing the amount of data requiring processing. The system can process images much faster, making it more suitable for real-time applications. Processing smaller data volumes requires less energy, making this model more energy-efficient. The model can be adapted to various environments and lighting conditions. By combining motion detection with face recognition methods, the model can enhance accuracy by using additional motion-related information, which helps eliminate errors associated with noise and complex background images. Additionally, it reduces false positives by responding only to genuinely moving objects, thereby minimizing the likelihood of incorrect face recognition. In the fourth chapter, «Software implementation of the face recognition model based on the background subtraction algorithm» Python programming language, along with OpenCV, TensorFlow, and Pickle libraries, was used to develop the application. These tools were employed to create system components that can reduce the load on the telecommunication network and improve the efficiency of face recognition and identification.