信号像素化下的多模态定位方法OA

This paper proposes an image-pixel-guided signal space and vision fusion localization algorithm(CDV),which innovatively transforms the simplistic numerical descriptions of individual signals into expres-sions for inter-signal numerical differences.By introducing the Signal Feature Analogy Index(FFAI)to construct a Signal Feature Pixel Map(SFPM)with multi-view representations,the algorithm achieves a profound transition from one-dimensional spatial signal intensity characteristics to two-dimensional pixelated descriptions.At the model construction level,the algorithm leverages the DPSE network to incorporate a contrastive learning mechanism so as to establish a localization model rich in multimodal feature information through the deep mining of similarity metrics and relative signal strength differences,thereby significantly enhancing the system's perception of complex environmental features.To further ensure the reliability of localization results,this paper employs a pair of monocular vision sensors to simulate a"quasi-binocular"visual effect,utilizes geometric constraints to perform visual space calibration on the signal-based results and achieve effective synergy between signal representation and visual observation.Validated by real-world tests with a 1-m sampling interval,the multimodal CDV method demonstrates a superior performance,achieving a precise Root Mean Square Error(RMSE)of only 1.100m within the Euclidean distance scale,with the localization error rate strictly controlled below 10%.