Search

Users of the digital image correlation method are faced with the problem of poor operability, low repeatability, and lack of standardized specifications for spraying speckles. To solve the problem, the research proposed a rock deformation measurement method that obviates the need to spray speckles. A local binary model was established by using the local binary pattern (LBP) operator based on deep texture features on rock surfaces. The resulting LBP digital speckle pattern can substitute artificial speckle patterns and demonstrates high quality and strong applicability. Based on the LBP digital speckle pattern, the target tracking algorithm was employed to achieve non-contact measurement of the dynamic displacements of rocks. The feasibility and effectiveness of the algorithm in practical application were verified by conducting shear tests on granite and siltstone. Test results show that the deformation characteristics in the displacement nephograms are in line with the measured data pertaining to rock fracturing and conform to the basic characteristics of the shear failure of rocks. The deformation measurement method based on surface texture information can realize non-contact displacement measurement of rocks under conditions without speckles: this obviates the influence of the quality of sprayed speckles on the accuracy of the measurement of deformation.

To solve the problem of low recognition rate of micro motion, this paper proposes a multilayer depth motion maps human action recognition method based on new projection strategy and energy homogeneous video segmentation. Firstly, the paper proposes a new projection strategy that projects the depth image into three orthogonal Cartesian planes, so as to retain more behavioral information. Secondly, considering that the image of multi-layer depth motion maps based on the whole video can reflect the whole motion information of the video but ignores a lot of detail information, the paper adopts a video segmentation method based on energy homogenization to divide an action video into multiple sub-video sequences, which can more sufficiently depict the detail information. Lastly, this paper uses a local binary pattern feature descriptor to describe the detail texture features of depth motion maps and adopts kernel extreme learning machine classifier to recognize actions. Experimental results on MSRAction3D and MSRGesture3D show that the proposed algorithm can achieve the accuracy of 94. 14% and 95. 67%, respectively, and has higher accuracy than the existing algorithms. [ABSTRACT FROM AUTHOR]