Assessing LV wall motion by frame-to-frame curvature matching and optical flow estimation

The focus of this paper is on the development of a novel cross correlation technique for shape matching between two similar objects. The shape profile is derived from the curvature of the object profile. The cross correlation technique is applied to the shape profile of the two image objects to evaluate their similarity. An application of the shape profile technique is presented in the context of pose detection of human models from image sequences or videos. To extract the feature points on the human model, a number of human motion templates are constructed with designated feature points on each of the human motion profiles. The human model feature extraction procedure is divided into two steps, including: (1) pre-process and construction of motion templates, and (2) extraction of feature points. The best matched template image pairs and best fitted feature points are identified by evaluating the cross correlation coefficient of the corresponding shape profiles. The proposed method is demonstrated by experimental results.

show abstract

“…Due to the quantization error in the silhouette extraction process, the window-based method [3] is adopted to smooth out the noise.…”

Section: A Shape Curvature Profilementioning

confidence: 99%