This paper proposes a multi-RAT Software Defined Network (SDN) assisted collision detection system designed to protect Vulnerable Road Users (VRU) by predicting road hazards. The system benefits from the Network Function Virtualization (NFV) and Software Defined Network (SDN). It includes a collision detection service, which collects specific messages from road users and processes them to detect collision risks. We propose a solution that allows vehicles and VRUs (such as pedestrians, cyclists, etc.) to smartly adapt the rate of the awareness messages. The frequency adaptation is performed under a certain number of constraints related to the risk level with the aim to improve trajectory prediction accuracy while decreasing unnecessary signalization on the network side and optimizing the energy consumption for the VRU side. Our simulation results suggest that frequency adaptation helps improve energy consumption and network signalization without impacting trajectory prediction accuracy and collision detection reliability.