Automatic Violence Detection Technique Using Mobinet

Deep learning holds great significance in machine learning since it effectively addresses a wide range of problems. The ability to identify violence from various video surveillance systems is one such crucial real-world application. Many crimes are occurring in numerous public spaces as a result of inadequate security. Many methods have been proposed to solve this specific problem, however they have drawbacks. Additionally, they are ineffective since they are dependent on certain conditions. Therefore, we presented an efficient automatic violence detection method for video datasets. We created a video dataset consisting of 1000 videos, half of which featured violent content and the other half did non-violent content. We used a Deep Neural Network technique called MobilNet for detecting violence from videos. Additionally, we employed a variety of deep learning and machine learning strategies to increase the precision. With the training data, the classification model showed an accuracy of 97.50% while with the test data, the accuracy was 95.60%. Performance evaluation results demonstrated that the suggested method identified violent content in videos successfully. The method for video violence detection performed better than many other methods already in use.