Vehicle Collision Prediction Using Bird's-Eye View Risk Modeling

The system employs a unified detection and prediction framework that handles multiple object types (vehicles, pedestrians, cyclists) through a single integrated apparatus. The risk level determination unit calculates collision risk levels for different object categories using the same sensor inputs, while the surrounding environment prediction unit generates future state information applicable to all object types. This multi-functional approach enables the system to predict collision risks with vulnerable road users without requiring separate dedicated systems for each object type.

The system performs preliminary processing by generating a simplified bird's eye view that integrates current state information, lane information, and future state information before inputting to the collision mode determination unit. The surrounding environment prediction unit pre-calculates future state information using estimation or deep learning algorithms, preparing processed data in advance. This preliminary action reduces the computational burden on the CNN by providing pre-processed, structured input data, thereby lowering real-time processing complexity while maintaining prediction accuracy.

The system extracts and integrates only the most critical information elements needed for collision prediction: current state information from sensors, lane information from the lane detection unit, and future state information from the surrounding environment prediction unit. By selectively extracting and integrating these essential components into a simplified bird's eye view, the system achieves comprehensive collision prediction accuracy while avoiding the computational overhead of processing all available sensor data. This extraction approach maintains prediction reliability by focusing on key predictive factors.