

Artificial intelligence and intelligent algorithms are increasingly widely used in intelligent transportation systems. Traditional methods rely on single sensor data, which is difficult to deal with intersections without signal lights and high-speed movement scenes. The Internet of vehicles technology provides a new idea for multi-source data fusion, but there are still deficiencies in early warning algorithms for official cars. In view of this, the paper proposes a collision warning algorithm based on Kalman network-bidirectional circular neural network, combining vehicle state estimation, trajectory prediction and circular-rectangular combined collision model, to realize multi-source data fusion through Internet of vehicle communication. Improved Kalman filter is used for state updating, and bidirectional temporal feature extraction is introduced to optimize the prediction accuracy. The Experiments show that the collision warning accuracy of the complete model is 97.83±0.62%, the false alarm rate is only 3.25±0.72%, and the trajectory prediction error and collision time calculation error are 0.42±0.11 m and 0.19±0.05 seconds, respectively. Compared to the unimproved model, the RMS error of the trajectory prediction decreased from 0.3 to 0.1, and the heading angle deviation at 12 hours decreased from 0.95 to 0.65. Therefore, this method significantly improves the collision warning performance of official cars at intersections, verifies the effectiveness of multi-model collaboration and data fusion of Internet of vehicles, and provides a reliable solution for the intelligent transportation system.