An improved table method for coded target identification with application to photogrammetric analysis of soil specimen during triaxial testing

Abstract Accurate and efficient recognition and identification of coded targets are of great importance in coded target-based photogrammetry. Recently, a deep learning-based method has been utilized to recognize the coded targets. Then, a table method has been developed to decode the coded targets, identify falsely identified coded targets, and recover missing coded targets. This method takes advantage of the geometric arrangement of the coded targets. In this paper, an improved table method has been developed to improve the coded targets recognition and identification results. Blob analysis, instead of deep learning, is utilized to recognize coded targets. Then, the RANSAC algorithm was utilized to identify falsely identified coded targets. Based on that, interpolation was performed on both the outside CT stripes and on membrane. Finally, the IDs of coded targets on the membrane are renumbered, which can increase the density of the coded targets on the membrane by three times. The effectiveness and accuracy of the proposed method are validated by implementing it into three-dimensional reconstruction of soil specimens during triaxial testing in geotechnical engineering. Experimental validation results indicate that the proposed method can achieve more accurate and more efficient coded target recognition and identification results.