The document describes a system for detecting abnormal driving behaviors using deep learning algorithms. It discusses three CNN models - Wide Group Densely Network, Wide Group Residual Densely Network, and Alternative Wide Group Residual Densely Network. The AWGRD Network is able to better identify behaviors by taking the superpositions of features from previous layers. The system works by generating an AWGRD model from training images, uploading a video, and using the model to monitor and detect behaviors in each video frame in real-time. Screenshots show examples of the system detecting a driver using a phone or radio. While accuracy is limited due to the small training dataset and laptop hardware, the system can currently detect behaviors correctly up to 80% of the time.