Neural-network–enhanced iterative attitude estimation for non-cooperative targets using LiDAR–thermal infrared fusion

A non-cooperative target relative attitude estimation error compensation strategy is studied in this paper to address the problem that the color point cloud iterative algorithm alone converges to a local optimal solution when estimating large angle rotation. Multi-sensor fusion data is utilized in the proposed method to compensate for errors via a neural-network-augmented iterative algorithm. The attitude rotation matrix can be estimated more accurately when the result of the color point cloud iterative algorithm is combined with the output of the proposed error compensation neural network. Simulations are conducted using self-constructed training dataset. Construction of LiDAR data and thermal infrared image data employs different simulation software. The effectiveness and superior of the neural-network-augmented iterative algorithm is verified through a comparison with existing algorithms.