3D surface reconstruction for lower limb prosthetic model using modified radon transform

• Main Author: Mohd. Sobani, Siti Syazalina
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: QH Natural history
• Online Access: http://eprints.utm.my/id/eprint/79534/1/SitiSyazalinaPFBME2017.pdf; http://eprints.utm.my/id/eprint/79534/
• Institution: Universiti Teknologi Malaysia

Computer vision has received increased attention for the research and innovation on three-dimensional surface reconstruction with aim to obtain accurate results. Although many researchers have come up with various novel solutions and feasibility of the findings, most require the use of sophisticated devices which is computationally expensive. Thus, a proper countermeasure is needed to resolve the reconstruction constraints and create an algorithm that is able to do considerably fast reconstruction by giving attention to devices equipped with appropriate specification, performance and practical affordability. This thesis describes the idea to realize three-dimensional surface of the residual limb models by adopting the technique of tomographic imaging coupled with the strategy based on multiple-views from a digital camera and a turntable. The surface of an object is reconstructed from uncalibrated two-dimensional image sequences of thirty-six different projections with the aid of Radon transform algorithm and shape-from-silhouette. The results show that the main objective to reconstruct three-dimensional surface of lower limb model has been successfully achieved with reasonable accuracy as the starting point to reconstruct three-dimensional surface and extract digital reading of an amputated lower limb model where the maximum percent error obtained from the computation is approximately 3.3 % for the height whilst 7.4%, 7.9% and 8.1% for the diameters at three specific heights of the objects. It can be concluded that the reconstruction of three-dimensional surface for the developed method is particularly dependent to the effects the silhouette generated where high contrast two-dimensional images contribute to higher accuracy of the silhouette extraction. The advantage of the concept presented in this thesis is that it can be done with simple experimental setup and the reconstruction of three-dimensional model neither involves expensive equipment nor require any service by an expert to handle sophisticated mechanical scanning system.