Early Pedestrian Movement Detection Using Smart Devices Based on Human Activity Recognition

Abstract

In the future, vulnerable road users (VRUs) such as cyclists and pedestrians will be equipped with smart devices capable of communicating with intelligent vehicles and infrastructure. This allows for cooperation between all traffic participants, such as cooperative intention detection and future trajectory prediction for advanced VRU protection. Smart devices can be used to detect the pedestrians’ intentions to warn approaching vehicles. In this article, we propose a method based on human activity recognition for early pedestrian movement transition detection using smart devices. These movement detections serve as valuable information for pedestrian path prediction and intention detection. We represent the pedestrians’ behavior using four states, i.e., waiting, starting, moving, and stopping. The movement transition detection is modeled as a classification problem and tackled by means of machine learning classifiers. The labels for training the classifier are obtained by evaluation of recorded high-precision head trajectories. We compare two different classification paradigms: A simple support-vector machine with linear kernel and a more complex XGBoost classifier. Our empirical studies with real-world data originating from experiments which 11 test subjects involving 79 different scenes show that we are able to detect movement transitions robust and early, reaching an F1-score of 85%.