An improved Kalman filter framework for pedestrian localization using radar, camera, and GPS fusion in IoV

Purpose In the Internet of Vehicles (IoV) environment, accurate localization of pedestrians and other road users is essential for intelligent transportation and safe mobility. Traditional single-sensor systems, such as radar or camera, are easily affected by line-of-sight and occlusion, limiting their performance in complex traffic scenarios. This study aims to develop a multi-sensor collaborative localization method to overcome these limitations. Design/methodology/approach A cooperative perception system integrating millimeter-wave radar, camera and pedestrian Global Positioning System data is constructed. To address heterogeneous observations between radar and camera, a dual-domain matching mechanism combining physical and image spaces is designed, with the Hungarian algorithm used for global optimal matching. Furthermore, an improved Extend Kalman filter, Mahalanobis Distance Weighted and Noise Adaptive Extend Kalman Filter (MWNA-EKF), is proposed to enhance robustness and accuracy. Findings Experimental results demonstrate that the proposed method achieves significantly higher fusion accuracy. Compared with the standard Extend Kalman filter, MWNA-EKF yields smaller localization errors in both lateral and longitudinal directions. Additional comparisons show that the three-sensor fusion scheme consistently out-performs any two-sensor combinations, confirming the effectiveness and superiority of the proposed approach under complex environments. Originality/value The main contribution of this study is the introduction of a dual-domain matching mechanism and the MWNA-EKF algorithm, which enhance the robustness and accuracy of multi-sensor fusion. More importantly, the proposed approach demonstrates clear application value by enabling reliable vehicle–pedestrian cooperative perception under complex traffic conditions. This provides a practical solution for improving localization accuracy and safety in intelligent transportation systems.