Riassunto analitico
Human activity recognition focuses on recognizing the action performed by a human. This kind of field has been employed in video surveillance systems to detect and analyze human activities for enhanced security. Additionally, in healthcare, human action recognition can aid in monitoring patient movements and activities, facilitating remote patient care. Moreover, in the field of robotics, accurate recognition of human actions enables robots to understand and respond to human gestures and commands, facilitating natural human-robot interaction. Another problem in this field is the scarcity of good datasets due to the needs of huge and diverse datasets.
This thesis addresses the problem of shoplifting detection, focusing specifically on robberies occurring in supermarkets, particularly through self-checkout systems. Shoplifters employ various clever techniques, such as hiding barcodes or not swiping them over the scanner, to deceive the system. Manual analysis of security videos for detection purposes is time-consuming, prompting the adoption of AI and Computer Vision technologies as efficient alternatives.
The research primarily targets three types of shoplifting: 1) hiding the barcode during scanning, 2) avoiding barcode swiping over the scanner, and 3) occluding the barcode with another item. Shoplifting detection is a relatively niche task, with limited academic literature and publicly available datasets. Consequently, many approaches rely on related tasks like hand gesture recognition. For this study, we utilize a dataset provided by Datalogic, a company specializing in barcode scanning technologies. The dataset consists of numerous frames capturing individuals repeatedly engaging in the afore mentioned shoplifting behaviors. Additionally, an initial portion of the research leverages the IPN Dataset as a baseline for preparing to tackle the actual problem. The ultimate objective is to compare state-of-the-art architectures specifically suited for this shoplifting detection task. The evaluation will employ various metrics to analyze and interpret the results obtained. By conducting this comparative analysis, the study aims to contribute to the advancement of efficient shoplifting detection methods.
|
