Interaction Behavior Recognition Between Intelligent Vehicles and Cyclists in Mixed Traffic

In mixed-traffic environments, the interaction between cyclists and vehicles is critical for vehicle safety and has become a key factor in traffic safety. To enable advanced driver assistance systems (ADAS) to effectively identify such interactions, a method for interaction behavior recognition based on graph representation learning was proposed. This method integrated the skeletal information of cyclists and key information of non-motor vehicles, modeled the spatiotemporal characteristics of cyclists, and incorporated key information from vehicles to extract the interaction features between cyclists and vehicles. Different labeling methods were proposed for behaviors with varying complexities. For simple and direct basic interaction behaviors, manual annotation was used to generate behavior labels to ensure data quality. For more complex and variable interaction behaviors, a graph kernel-based clustering algorithm was employed to automatically generate behavior labels, addressing behaviors that were difficult to classify manually. A cyclist-vehicle interaction behavior graph model dataset was constructed based on the cyclist spatiotemporal graph model and behavior labels, and the recognition of interaction behaviors was accomplished using a graph kernel-based classification method. Real-vehicle data collection experiments were conducted to validate the effectiveness of this method. In the experiments, the system collected a large amount of actual cyclist-vehicle interaction behavior data, and behavior recognition was performed using the proposed method. The experimental results demonstrate that this method can identify various interaction behaviors between cyclists and vehicles with an accuracy of 99.65%. This high accuracy indicates that the method has significant practical value for improving the safety performance of intelligent driving systems, providing theoretical and practical support for further advancements in intelligent onboard systems.