Real-Time Collision Detection Algorithm with Redundancy Based on Intention and Trajectory Prediction

In the field of autonomous driving, trajectory prediction and collision detection have been extensively researched, yielding many impressive algorithms. However, many of these algorithms encountered challenges when applied in real-world scenarios due to their limited real-time performance and inadequate perception of potential hazards. To address these issues, we propose a long-term trajectory prediction algorithm frame with great real-time performance. In this algorithm, two different models based on (Long Short Term Memory) LSTM were employed to predict driving intentions and trajectories with an 8-s duration for road and intersection scenarios. Moreover, we introduce spatial and time redundancy to the conventional oriented bounding box (OBB) based collision detection algorithm to enhance reliability. According to the simulation results in SUMO, the proposed model meets the engineering requirements for trajectory prediction and collision detection regarding real-time performance and reliability.