Researchers at Xiamen University’s School of Information, led by Professor Wang Hanzi, have advanced robust model fitting in computer vision, publishing their findings in IEEE Transactions on Pattern Analysis and Machine Intelligence. The study addresses longstanding limitations in traditional and learning-based approaches to estimating model parameters from noisy, outlier-corrupted data. Conventional methods like RANSAC suffer from low computational efficiency and fuzzy model hypotheses due to reliance on iterative sampling and neglect of data preference distributions. Meanwhile, existing deep learning techniques are constrained by static coordinate representations and the limited receptive fields of convolutional neural networks, hindering their ability to capture complex motion cues. To overcome these challenges, the team developed MPCFormer, a motion-aware pyramid transformer that eliminates iterative sampling by integrating correspondence learning with spatiotemporal motion cues. The architecture reframes robust model fitting into a joint optimization framework for point classification and parameter estimation. Central to the method is a motion preference filter, which employs transformer blocks and multi-head attention to explicitly encode preference distributions and adaptively suppress outliers. Complementing this is a pyramid consensus smoother that builds a sparse feature pyramid via multi-scale transformer encoding, capturing hierarchical motion consistency from local to global scales to resolve spatial discontinuities. Experimental validation demonstrates significant performance gains over state-of-the-art baselines, with improvements of 4.68 percent in mAP@5°, 1.89 percent in AUC@3 pixels, and 1.52 percent in F-score. The approach establishes a more efficient and accurate pathway for motion structure recovery in intelligent vision systems. The research was conducted by PhD student Yin Wenyu under the joint supervision of Wang Hanzi, Associate Professor Lin Shuyuan of Jinan University, Professor Yan Yan, Associate Professor Lu Yang, and Professor David Suter of Edith Cowan University. The project received funding from the National Natural Science Foundation of China.