Fight detection
For the detection of fights, we are using a combination of techniques based on supervised learning, anomaly detection, and structured deep learning achieving state-of-the-art solutions which work in real-time, generalizing diverse types of violence both indoors and outdoors.
Again, the results are obtained in real time, achieving accuracies above the state of the art, and marking the temporal instants in which violence is found, so that these significant video segments can be provided as evidence to the police if needed. In addition, human in the loop can be included to monitor and corroborate all cases but narrowing its work to those cases where there are indications of violence, without the need to monitor/supervise multiple screens at the same time.