Electric Scooter and Its Rider Detection Framework Based on Deep Learning for Supporting Scooter-Related Injury Emergency Services

Nguyen, Hoa; Nguyen, Minh; Sun, Qian

Electric scooters (e-scooters) have been considered as a “last mile” solution to existing public transportation systems in many cities all over the world due to their convenience at a highly affordable price. E-scooters enable users to travel distances which are too long to walk but too short to drive, so they help to reduce the number of cars on the roads. Along with its increasing popularity, accidents involving e-scooters have become a growing public concern, especially in large cities with heavy traffic. It is useful to detect and include e-scooters in traffic control. However, there is no available pre-built-model for detecting an electric scooter. Therefore, in this paper, we proposed a scooter and its rider detection framework that supports emergency management for scooter-related injuries. The framework helps to identify scooter and its rider in live-stream videos and can be applied in traffic incidents detection applications. Our model was developed based on deep learning object detection models. Using ImageAI API, we trained and deployed our own model based on 200 images acquired on the internet. The preliminary results appeared to be robust and fast; however, the accuracy of our proposed model could be improved if using a larger dataset for training and evaluating.