A Fast Image Matching Approach in Fatigue Testing of Wall Plate Structures

To address the challenges in motion vector estimation due to local non-uniform deformation of specimens during fatigue tests, this paper proposes an image matching method that combines template sub-image techniques with a Kalman filter. This method employs a mesh motion estimation model, initially using the Kalman filter to predict the matching point positions of mesh nodes. Subsequently, template matching is utilized to achieve precise matching of these nodes. The algorithm is accelerated through GPU-CPU heterogeneous parallel computing technology. Test results indicate that the GPU-accelerated image matching method attains the accuracy of the CPU-based algorithm while significantly reducing processing time, thereby fully meeting the real-time requirements of online image feature matching applications.