Multi-Modality Fusion Perception Strategy Based on Adaptive Matching for Vehicle-Road Cooperation

With the continuous development of autonomous driving technology, new requirements have been put forward for the perception performance of traffic scenes, and vehicle-road cooperative technology has attracted the attention of the industry. In the face of increasingly complex traffic scenes and huge interactive perception data, how to balance the detection accuracy and real-time requirements of the vehicle-road cooperative perception system has become an urgent problem to be solved. In this paper, an adaptive selection method based on heterogeneous multimodality perception data and processing model optimization method is proposed to evaluate and optimize the quality of multi-end sensor data, and optimize the spatial scale of feature extraction model according to the complexity of the perception scene environment, so as to realize accurate real-time perception of vehicle-road collaboration. At the same time, a multi-spatial scale modulation module is introduced for the characteristics of scenes with different environmental complexity, which effectively integrates local and nonlocal feature information and enhances the ability of the model to process features in the scene of cooperative vehicle and road perception. The experimental results compare and analyze the accuracy and time delay performance of feature extraction models with different scales of multimodality data. The performance of the adaptive selection multimodality fusion perception strategy proposed in this paper is compared and analyzed. The results show that the proposed fusion strategy has 9.87% improvement in comprehensive detection performance compared with the optimal detection scheme, and 27.58^% improvement for the shortest delay scheme. At the same time, the multi-scale spatial modulation method for feature extraction of sensory data is significantly enhanced at the visualization level.