An Optimized GNSS RTK/INS/Vision Integration-Based Vehicle Positioning Model and Its Credibility Assessment

Accurate positioning is critical to Intelligent Transportation Systems (ITSs). Current research primarily focuses on improving Global Navigation Satellite System (GNSS) positioning accuracy and continuity through multi-sensor integration. With the emergence of new industries such as assisted driving, the credibility of positioning results has gradually attracted attention. To explore the feasibility of achieving credible positioning, this paper presents an optimized Inertial Measurement Unit (IMU) and camera tightly augmented GNSS Real Time Kinematic (RTK) model, along with the positioning credibility assessment. In this model, multi-factors that affect the positioning errors are considered as the feature inputs of the Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) network, then a credible factor and its uncertainty are generated. Moreover, a positioning optimization algorithm is presented based on the credible factor. To evaluate the effectiveness of the presented model, several sets of vehicle-borne data in urban environments are processed and analyzed. Results illustrate that (1) the presented positioning model achieves comparable positioning and superior orientation determination accuracy compared to existing state-of-the-art methods; (2) the generated credible factor can envelop 94% horizontal positioning errors and 84% vertical positioning errors with envelope levels of 5 cm and 7cm; (3) the error coverage rate of confidence interval generated by the uncertainty of credible factor in horizontal and vertical directions can reach 94% and 87.57%, which is close to the theoretically set 95% confidence interval for horizontal direction; (4) the positioning results can be optimized while applying the position optimization algorithm based on credible factor.