A New Framework for Video-based Frequent Iris Movement Analysis towards Anomaly Observer Detection

This paper suggested a new framework for detecting abnormal behavior, specifically based on frequent iris movements. It contributed to a decision whereas an individual is dubious or unsuspected from a video. One of the key components of questionable observer detection is to detect some specific suspicious activity. According to the writer, various areas of the body movement and human behaviors may be an indicator of suspicious behavior. In this research, we considered the movement of human eyes to identify suspicious activity. This working field is also a significant aspect of machine vision and artificial intelligence, and a big part of the understanding of human behavior. The system framework comprises three parts to monitor suspicious video activities. First, we used the Multi-task Cascaded Convolutional Networks (MTCNN) classifier to detect eyes. Second, we observe irises from eye representations with the use of Circular Hough Transformation (CHT). Finally, we calculated the average distance of iris movement from eye images using a new morphological method called TRM using some properties of the iris movement. We have observed a particular phenomenon of frequent iris movement. Hence, we are making a case of someone being an abnormal person and referring it to a suspicious observer. To vouch for our work, we created our data set with 100 videos where 30 individuals volunteered to validate this research. Each video comprises 200 frames with a duration of 6-10 seconds. We’ve reached an accuracy of 94% on detecting a frequent iris movement. Rather the goal is to minimize people’s burdens so they can focus on a small range of cases for investigation in more depth. This research’s sole purpose is to indicate a person’s anomalous behavior on the basis of frequent iris movement. Our research outstrips much of the current literature on abnormal iris movement and dubious investigator identification.