The use of artificial intelligence is a practice that is increasingly integrated into video surveillance systems, whether closed-circuit television systems or systems with intelligent IP cameras. In this way, video surveillance systems usually offer, at low cost, different functionalities such as object and person detection, facial recognition, issuing intrusion alerts, etc. However, with this dynamic has also increased the need to store, transmit and manipulate large amounts of data associated with video and the additional information that is generated daily. In this article we present a methodology and a video surveillance application focused on the automatic detection of abnormalities in pedestrian flow, using smart IP cameras, pre-trained YOLOv8 and statistical event recording with MongoDB. The accuracy tests in counting and tracking pedestrian flow that we have carried out give us results of Precision = 90%, Recall = 84%, Specificity = 62.5% and F1 = 87%, in addition to generating alerts when it detects abnormal pedestrian flows.