This project envisions a system that can be used for two use cases: Fight Detection on Campus and Drowsy Driver Detection. The scope of our current project is limited to detecting small fights occurring on campus and alerting the driver when they start to feel drowsy. We limit our current applications to a single camera usually remote, which needs to be connected to our application and inference models. We propose a website hosted on a server to streamline this process.

RCN (Long-term Recurrent Convolutional Networks) combines CNNs and RNNs to recognize human activities in videos. LRCN first extracts visual features from each frame of the input video using CNNs and then processes these features with an RNN, such as an LSTM or GRU, to capture the temporal dynamics. The RNN component allows LRCN to model long-term dependencies between frames, which is crucial for activity recognition. By integrating CNNs and RNNs, LRCN can predict activity labels for each video segment and localize the temporal boundaries of the activity.

• Flask 2.2.3 and above
• PyQt5 5.15.1 and above
• OpenCV 3 or 4
• Tensorflow 2.11 and above
• CUDNN 8.4 and above
• CUDA 11.1 and above

Drowsy Driver Detection Fight Detection