Abstract:

For the problem that current multi-sensor information fusion in intelligent vehicle environmental perception systems often involves phased fusion of different sensors， making it difficult to balance the accuracy advantages of individual sensors and the redundancy advantages of multi-source information， an object-level parallel-structured unified multi-sensor information fusion method based on the interacting multiple model （IMM） is proposed in the paper. Object-level fusion has excellent modularity and encapsulation. The parallel structure can fully utilize information redundancy advantages， and the interacting multiple models enable unified and efficient fusion of multi-source data， compensating for the limitation of individual sensors. After spatiotemporal alignment of multi-source sensor data， the nearest neighbor method and DS evidence theory are used to achieve multi-sensor information association， and then dynamic unified fusion based on the interacting multiple models is conducted. Real-vehicle experiments are conducted using millimeter-wave radar and a vision system for environment perception. The results show that the proposed method effectively improves the reliability and stability of target vehicle perception and tracking， enhancing the adaptability of the system.

Key words: intelligent vehicle, environment perception, object-level fusion, parallel filtering, interacting multiple models