A Framework on Deep Learning-Based Indoor Child Exploitation Alert System

Child violence is one of the most heinous crimes prevailing in our society. Any form of abuse or violence to a child does matter and cannot be overlooked. It affects the mental health of a child so deeply that it influences his later life. So, taking proper measurements for saving every child from any sort of violence is a must. This paper proposes a modified deep learning-based violence detection framework that will be able to detect and alert child exploitation in real-time without any privacy breach. The proposed framework is designed to work indoor, so primarily it can be installed in homes and educational institutions. The mechanism of the system is as such firstly, the surveillance video streams will be optimized using a modified convolutional neural network (CNN). Secondly, a sequence of frames will be passed through CNN for feature extraction and transferred to the long short-term memory (LSTM), which will act as a classifier. A softmax layer also has been introduced for the probabilistic distribution. Finally, the age and activities of a specific person will be detected. If there is any violent activity, an alert will be sent to the guardian through the system. We intend to ensure our proposed framework will be implied for automatic violence detection in a quick and safe approach.