Video based Anomaly Detection Utilizing the Crow Search Algorithm-based Deep RNN

Abstract

Numerous facets of our life have changed recently as a result of deep learning. The advancements in artificial intelligence enable computers to perform more of our regular chores. These days, there are more disruptive and provocative actions taking place than ever before. Security has thus been given priority consideration. CCTVs are being installed in more public locations, such as shopping malls, streets, banks, etc., to ensure people's safety. Because of this difficulty, a very accurate computerization of this system is now necessary. Since it would be very difficult for people to continuously monitor these security cameras. To determine whether the recorded actions are aberrant or suspicious, it demands workforces and their continual attention. Consequently, this shortcoming is driving a demand for highly accurate automation of this operation. The paper discusses the deep learning implementation technology that underlies the different crowd video analysis methods. For a variety of factors, including simplicity, performance, computational effectiveness, and high-quality interpretability, feature selection is crucial. Due to all the significance outlined above, an unique feature selection technique for anomaly detection that combines RNN and the crow search algorithm is proposed (CSA-RNN). In order to maximise the benefits of global search, the best attributes should be taken into account in each iteration.  Additionally, it is necessary to identify which frames and sections of the recording include the odd activity in order to make a quicker determination of whether the unusual behaviour is suspicious or atypical. Its goal is to inevitably identify aggressive and violent behaviours in real time while removing variations from expected patterns. To recognize and categorize different levels of high movement in the frame, we wish to use a variety of Deep Learning model RNNs. From there, we may send out a danger detection alert, warning individuals of any ominous behaviour at a certain time.