As part of this project, We aimed to explore the potential of using computer vision and machine learning (ML) to detect real-time violence. We utilized various techniques such as object detection and activity recognition algorithms, as well as facial recognition, to analyze visual data captured by cameras and other imaging devices in real-time.

One of the approaches that we took was to train object detection algorithms to recognize and locate specific objects or people in an image or video. For example, the algorithm was trained to recognize and locate a person holding a weapon, or to detect signs of physical altercations such as pushing or hitting. This would allow for the detection of potentially violent behavior in real-time.

Another approach that we took was to use activity recognition algorithms to recognize specific actions or behaviors. For example, the algorithm was trained to recognize and classify behavior such as fighting, or to detect when someone is pointing a weapon at another person.

Additionally, We also used facial recognition algorithms that were trained to recognize facial expressions that indicate anger or aggression. This added an extra layer of detection for potentially violent behavior.

As a crucial part of the project, We also considered ethical and privacy concerns when using computer vision and ML to detect real-time violence. We ensured that the algorithm was not biased against certain groups of people and that individuals were not being surveilled without their consent.

Overall, this project highlighted the potential of using computer vision and ML to detect real-time violence, but it is important to note that these technologies are not perfect and need to be developed and implemented in an ethical and responsible manner.

We plan to expand upon this by training the model on sports footage so that we can train an LLM to provide automated commentary.