Real-Time 5-DOF Object Pose Estimation Using Uniaxial Magnetic Sensor Array and MPNet Convolutional Network

With advancements in intelligent technologies, near-field pose estimation has become essential across various applications. This paper presents a 5-degree-of-freedom (DOF) real-time object pose estimation system based on a uniaxial magnetic sensor array and MPNet convolutional network. A simulated dataset of magnetic field distributions is created for training, utilizing a data augmentation method that reduces pose space traversal by up to 98.43%. The SMPNet network, built on a ResNet backbone, achieves high classification accuracy and Pearson correlation for object pose estimation. A multi-channel data acquisition system with an 8 ×8 sensor array is designed, and real-world data are used to optimize the model, achieving a mean square error of 2.02 mm for XYZ displacement and 5.92[Formula: see text]for rotational angles. The system is integrated with hardware to minimize data transmission delays, achieving a total system power consumption of 2.058 W and an estimated pose RMS error of 2.38 mm and 7.56[Formula: see text]. This study demonstrates the feasibility of using a uniaxial magnetic array for real-time pose estimation, offering insights for future system optimizations and practical applications.