Intelligent video stream object recognition system using machine learning techniques
Students Name: Rudchyk Illia Valentynovych
Qualification Level: master (ESP)
Speciality: System Analysis
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2021-2022 н.р.
Language of Defence: ukrainian
Abstract: Comprehensive optimization of production processes and technological operations using computer vision technology is a feature of modern society. At the same time, among the common approaches to recognition are the following areas based on optical pattern recognition (used mainly in OCR systems), methods of circuit identification (contour analysis algorithms) and the use of artificial neural networks and machine learning technologies [1]. The task of object recognition consists of two parts: learning and recognition. Training is done by displaying independent objects of one class. As a result of the training, the system must be able to respond to all objects of one class. Recognition process that determines the actions of an already trained system goes next. Informatization of this process is the main problem. Leading global companies such as Google, Microsoft, Facebook Apple, Intel have set up departments to develop image recognition libraries, but their results boil down to applications for recognizing animals, humans, and etc [2]. The analysis of the existing software (VLC Media Player, MotionDetect detector, Table View Video Player, dvr-scan, Yawcam) showed their narrow orientation and many disadvantages (commercialization of use, limited functionality, poor quality of recognition). As a result, the urgent task is to create an object recognition information system in the video stream Today, scientific sources widely present various methods of object recognition and computer vision. In particular, Ajeet RamPathak paper[3] demystifies the role of deep learning techniques based on convolutional neural network for object detection. Deep learning frameworks and services available for object detection are also enunciated. Very wide overview of past research on object detection, outline the current main research directions, and discuss open problems and possible future directions described in "Object Detection: Current and Future Directions" [4]. Study object : the process of recognizing moving objects in a video stream. Scope of research: methods and means of object recognition in the video stream using machine learning methods Goal of research: to develop an intelligent system that is designed to recognize moving objects in the video stream using machine learning while saving the relevant sections of video in a file for human control and classification of certain moving objects. The algorithm of the software is as follows. At the beginning of the video, the flow is analyzed for any movement on which noise cancellation (wind fluctuations, slight shifts) is performed. When motion is detected, the object is framed and tracked with further grading. On this step machine learning modules performs several algorithms to detect to what type the object is referred. Selected contours of objects are compared with a large set of data objects that have already been identified and at a certain threshold of convergence, decisions will be developed on the classification of an object[5]. The result of the work is the original image which with high accuracy can determine whether the object belongs to a particular cluster Keywords: Computer vision, machine learning, object recognition, motion detection, video processing, noise cancellation. References 1. Tufte, E. (2001) The Visual Display of Quantitative Information / E. Tufte. Second edi-tion. Connecticut: Graphics Press, 206p. 2. Dix, A.(2009) Human-Computer Interaction / A. Dix. New York, USA: Springer US, P. 1327–1331. 3. OpenCV (C++ vs Python) vs MATLAB for Computer Vision [Electronic source] / Ac-cess mode: https://www.learnopencv.com/opencv-c-vs-python-vs-matlab-for-computer-vision. 4.Application of Deep Learning for Object Detection, 2018. URL: https://www.sciencedirect.com/science/article/pii/S1877050918308767 – 9.05.2018р. (дата звернення: 21.05.2022). vision. 5.Object Detection: Current and Future Directions , 2015. URL: https://www.frontiersin.org/articles/10.3389/frobt.2015.00029/full – 17.11.2015р. (дата звернення: 21.05.2022).