Fisheye cameras provide an expansive field of view that surpasses 180° and are frequently employed in high-security environments, including retail establishments and parking facilities. Nevertheless, these cameras pose significant obstacles for conventional surveillance frameworks, since the inherent distortion in their images complicates the identification of atypical behaviors such as shoplifting or vehicle theft, necessitating human oversight. Current systems frequently encounter difficulties in effectively recognizing such occurrences. To answer this question, we present a study in several algorithms that could help in the detection of anomalous behaviors of fisheye images. Our approach entails a new deep learning methodology wherein automatic and accurate detection will be made possible through the application of convolutional neural networks. We apply the YOLOv8 model with Xception as our base model to identify abnormal activities correctly and locate frames in which suspicious actions occur. This method enhances crime deterrence through instantaneous detection and, in addition, can act as a cost-effective alternative to current monitoring methods, thereby significantly reducing reliance on labor-intensive monitoring