THEFT DETECTION WITH CRIMINAL IDENTIFICATION USING MACHINE LEARNING

Abstract [English]

One of the main goals of video surveillance research and practical implementations is abnormal event detection. In order to improve public safety, the usage of surveillance cameras in public spaces—such as roadways, crosswalks, banks, retail centers, etc.—is expanding. One of the most important tasks in video surveillance is the detection of anomalous occurrences, such as criminal activity, traffic accidents, or crimes. In general, abnormal events are rare in comparison to normal activities. A useful anomaly detection system aims to pinpoint the anomaly's temporal range and instantly notify users when any behavior deviates from expected norms. Consequently, it is possible to think of anomaly identification as coarse-grained video knowledge that separates anomalies from regular patterns. Classification techniques can be used to further categories an anomaly into one of the specific activities once it has been recognized. An overview of anomaly detection is provided in this work, with a particular emphasis on applications in banking operations. Banking operations involve a wide range of daily, weekly, and monthly tasks and exchanges carried out by or impacting several parties, including staff members, clients, debtors, and outside organizations. Events could develop gradually, and early identification greatly reduces the likelihood of negative consequences and, in certain situations, even completely prevents them. Finding people at unfavorable periods is accomplished using anomaly detection based on time series. This research offers a machine learning based anomaly detection technique to discriminate between normal and abnormal occurrences. A comparison is made between the biometric identity of the captured face and the biometric identities of known criminals. If a match is found, we can identify and capture the culprit right away.