Enhancing Pedestrian Safety: Predicting Movements with Deep Learning Models

In the USA, road accidents are an issue of concern, especially regarding pedestrian uncertainties. It is challenging for many drivers to take sudden action visualizing the movement of pedestrians. Predicting pedestrian motion and generating automatic machine responses, such as slowing the car or alerting the driver, can help reduce these unwanted incidents and save many lives. To address this, we used the MotionSense dataset, comprising six activity measurements from 24 subjects, to predict their activities sequentially. Three deep learning approaches have been considered in this study, which are a hybrid model of long short-term memory (LSTM) and a two-dimensional convolutional neural networks (2D CNN) model (LSTM-2D CNN), an LSTM model, and a CNN model. The model is expected to identify the activity from the given feature. The outcome of these three models is that they can predict the activities with over 90% accuracy. The LSTM model achieved the highest accuracy, approximately 93%.