<p>Detection of Vulnerable Road Users (VRU), such as pedestrians, cyclists, is a crucial task for autonomous driving to plan evasive maneuvers. Most of the current VRU intention recognition algorithms focus mainly on pedestrians typically using computationally intensive resources, and also the extension to other classes requires significant effort. We propose a VRU intention recognition algorithm that reduces the complexity and computational resources required for the task by using the bird’s eye view from the perspective of the camera mounted on the ego vehicle. Our main objective is to show the importance of mapping the positions of all VRU in a scene onto a 3D world coordinate system for getting simplifications and accelerations in the pipeline. The approach for pedestrian is scalable and easily deployed for the cyclist intention recognition task as well. Testing is done on the JAAD dataset for pedestrians and the CASR dataset for cyclists. Compared to the existing methods, our framework is light weight and uses significantly lesser number of parameters and has a faster training and inference time with comparable accuracy. Additionally, we also propose a methodology to reduce the gap in the data distribution between real and synthetic data generated by simulation for the VRU intention recognition task. The synthetic data generation improved the F1, area under the curve (AUC) and precision metrics. Overall our method aims at striking a balance between different relevant metrics for VRU intention recognition.</p>