Federated Learning for Physical Violence Detection in Videos

Domestic violence has increased globally as the COVID-19 pandemic combines with economic and social stresses. Some works have used traditional feature extractors to identify body positions to detect physical violence. Besides, the use of Machine Learning is limited by the trade-off between collecting more data while keeping users' privacy. Federated Learning (FL) is a technique that allows the creation of client-server networks, in which anonymized training data can be uploaded to a central model, responsible for aggregating and keeping the model up to date, and then distributing the updated model to the clients' nodes. This paper proposed an FL approach to the violence detection problem in video. The framework was evaluated on AIRTLab Dataset, in which frames were extracted. It used pretrained Convolutional Neural Networks (CNN) to address the image classification problem. Inception v3, MobileNet v2, ResNet152 v2, and VGG-16 architectures were evaluated, with the MobileNet architecture presenting the best performance, in terms of accuracy (99.4%), with a loss of 0.5% when compared to the non-FL setting.