Search

Purpose. Research on the dependence of the amount of microcontroller memory used and the image recognition time on the type of convolutional neural network MobileNet V1 or V2 and their hyperparameters. Methodology. Creating a database of images necessary for training a neural network using the Edge Impulse software platform with subsequent loading of the network into the memory of a 32-bit microcontroller ESP32 for practical evaluation of network characteristics. Findings. A comparison of the characteristics of MobileNetV1 and MobileNetV2 neural networks was made. Experiments were conducted to determine the dependence of the recognition time of selected objects on the image, the amount of used RAM and program memory, based on the ESP-EYE microcontroller with a camera, depending on the network width ratio, image size and convolutional neural architecture network. It was determined that the time of image classification for the MobileNetV2 network model takes from three to ten seconds or more, which is not acceptable for fast recognition tasks. It was also found that the microcontroller program memory is insufficient to analyze the 160 by 160 pixel images with the maximum network width of the MobileNetV2 model. Using the MobileNetV1 network provides slightly lower recognition accuracy, but requires significantly less microcontroller resources and time. Originality. The peculiarities of the use, possibilities and limitations of neural networks for image recognition in systems based on ESP32 microcontrollers have been verified in practice. The dependence of the amount of memory used on the microcontroller and the time of image recognition on the type of convolutional neuron was established in order to select additional image processing tools to improve the quality of recognition. Practical value. The obtained results allow the proper selection of the MobileNetV1 or MobileNetV2 neural network depending on the specific tasks of image recognition by systems on microcontrollers. [ABSTRACT FROM AUTHOR]

The history analysis of video surveillance systems, as well as models, methods and technical means used in modern systems were carried out and was proposed new approaches to improve them. A review of existing solutions in the field of video surveillance systems was carried out, namely: hardware, functions, structure and features of modern systems. Modern models and methods that can be used for video analytics in video surveillance systems are analyzed, such as: anomalies detection, deep neural networks for classification, localization, segmentation, detection, identification and tracking of objects in the video. Nowadays, video surveillance has become widespread due to the rapid development of both hardware and computer vision, methods of artificial intelligence, in particular, machine learning based on deep models. These achievements can be used for better video analytics and thus reduces the number of false alarms, as in cases where important objects and events are not detected during video surveillance, and vice versa, when false alarms are given. Therefore, modern video surveillance systems need to be modernized using new possibilities. It is proposed to use the YOLO convolutional neural network to analyze the video in order to search for dangerous / suspicious events; to reduce the load on the server from the use of the neural network, the system will analyze the change of the current video frame compared to the previous one and use models and methods to detect anomalies, as well as the principle of locality. As a result of research, the requirements for security systems based on video surveillance and approaches to their development are proposed. The client-server structure and functions of the system are designed. Two initial versions of the video surveillance system have been developed. On this basis it is possible to implement various solutions and conduct experiments. The applied equipment, tested software, results of work are described. [ABSTRACT FROM AUTHOR]

Successful identification of a person in a real-time video stream is a critical function in a wide range of relevant subject areas. On the other hand, such a task is complex and can be effectively solved only with the help of modern artificial intelligence approaches. In this work, the previously developed MTCNN approach is used as a key algorithm for face image recognition and vectorization, which uses a cascade of several neural networks to identify key points of the face of a recognizable personality. In addition to the main algorithm, an integrated system was implemented that provides the functionality of person identification available to the end user. Additionally, face identification modules were implemented based on vectors representing a person in a video frame, a subsystem for storing and replenishing the database of recognizable persons, and a convenient user interface. Modifications of the MTCNN architecture proposed in this paper made it possible to achieve real-time performance and acceptable identification quality at the level of 0.92 according to the F1-metric. [ABSTRACT FROM AUTHOR]