An Improved Magnetic Tracking Approach Based on ResNet-LM Fusion Algorithm
- 25 October 2022
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-10
Abstract
Permanent magnet-based tracking (PMT) approach is a reliable solution for motion tracking and navigation. The conventional PMT for tracking one or more magnets is based on optimization algorithms such as Levenberg-Marquardt (LM) algorithm. However, the tracking accuracy and computation time of the LM algorithm depends on the initial values of pose parameters. The artificial neural network (ANN) based methods can provide an alternative solution for optimization algorithms, whereas the studies based on machine learning algorithms had limited tracking performance. In this study, we proposed a ResNet-LM algorithm, based on the fusion of the deep learning algorithm and optimization algorithm, to improve PMT performance. Firstly, we employ a residual neural network (ResNet-20) to track a single magnet. The tracking accuracy of ResNet is (1.69±1.05 mm, 1.25±0.68°) by padding with a scale of 2, where the size of the magnetic flux density (XT) becomes (3, 8, 8). Secondly, the prediction result of ResNet is adopted as the initial value for the LM algorithm, with the improved tracking accuracy of (1.25±0.90 mm, 2.80±1.94°). Finally, we utilize an adaptive weighted fusion algorithm to fuse the prediction results of ResNet and LM algorithms. Experimental results show that the pose tracking accuracy of ResNet-LM is improved to (0.90±0.75 mm, 1.51±0.87°), where the mean computer time for the ResNet-LM algorithm was 79.3 ms. The ResNet-LM-based pose accuracy for AGV Parking was (6.09±1.18 mm, 2.34±1.23°). The proposed approach can improve the tracking accuracy and reduce the computational time of PMT, thereby extending the application scenarios of PMT.Keywords
Funding Information
- National Natural Science Foundation of China (61973293)
- Chinese Academy of Sciences through the Key Project of Foreign Cooperation for the International Partner Program (121835KYSB20190069)
- Central Guidance on Local Science and Technology Development Fund of Fujian Province (2021L3047, 2020L302)
This publication has 29 references indexed in Scilit:
- Accurate Indoor Location Tracking Exploiting Ultrasonic ReflectionsIEEE Sensors Journal, 2016
- An Active RFID Tag-Enabled Locating Approach With Multipath Effect Elimination in AGVIEEE Transactions on Automation Science and Engineering, 2016
- Locating Intra-Body Capsule Object by Three-Magnet Sensing SystemIEEE Sensors Journal, 2016
- Convolutional neural networks at constrained time cost2015
- A Simulated Annealing-Based Multiobjective Optimization Algorithm: AMOSAIEEE Transactions on Evolutionary Computation, 2008
- Multistrategy ensemble learning: reducing error by combining ensemble learning techniquesIEEE Transactions on Knowledge and Data Engineering, 2004
- Tracking system with five degrees of freedom using a 2D-array of Hall sensors and a permanent magnetSensors and Actuators A: Physical, 2001
- Genetic drift in genetic algorithm selection schemesIEEE Transactions on Evolutionary Computation, 1999
- Ant colony system: a cooperative learning approach to the traveling salesman problemIEEE Transactions on Evolutionary Computation, 1997
- Multilayer feedforward networks are universal approximatorsNeural Networks, 1989