Chengjiang Long

Transmission: a new feature for computer vision based smoke detection

A novel and effective approach is proposed in this paper to detect smoke using transmission from image or video frame. Inspired by the airlight-albedo ambiguity model, we introduce the concept of transmission as a new essential feature of smoke, which is employed to detect the smoke and also determine its corresponding thickness distribution. First, we define an optical model for smoke based on the airlight-albedo ambiguity model. Second, we estimate the preliminary smoke transmission using dark channel prior and then refine the result through soft matting algorithm. Finally, we use transmission to detect smoke region by thresholding and obtain detailed information about the distribution of smoke thickness through mapping transmissions of the smoke region into a gray image. Our method has been tested on real images with smoke. Compared with the existing methods, experimental results have proved the better efficiency of transmission in smoke detection.

Local binary pattern based texture analysis for visual fire recognition

Color and shape features have difficulty to recognize fire from fire-colored, irregular-shaped and moving objects. An effective algorithm using texture analysis with the Local Binary Pattern (LBP) is proposed in our paper to help deal with this problem. Besides grayscale and rotation invariance, the ability of LBP operator for multi-resolution analysis is enhanced through multiple LBP operators with varying parameters and pyramid based image sub-sampling. Experiments are performed on sample pictures of red fire, car and banner taking Mahalanobis distance classifier as the evaluation criterion. From experimental results the principles are analyzed for the selections of fire region, color channel and histogram equalization. To illustrate the efficiency of our adopted LBP method, its performance has been compared with those from Gray Level Co-occurrence Matrices (GLCM) and wavelet transform (WT).

A new region growing algorithm for triangular mesh recovery from scattered 3D points

A novel region growing algorithm is proposed for triangular mesh recovery from scattered 3D points. In our method, the new principle is used to determine the seed triangle considering both maximum angle and minimum length; the open influence region is defined for the active edge under processing; positional element is added into the criterion to choose the most suitable active point; geometric integrity is maintained by analyzing different situations of the selected active point and their corresponding treatments. Our approach has been tested with various unorganized point clouds, and the experimental results proved its efficiency in both accuracy and speed. Compared with the existing similar techniques, our algorithm has the ability to recover triangular meshes while preserving better topological coherence with the original 3D points.

An automatic method of measuring foot girths for custom footwear using local RBF implicit surfaces

Three-dimensional point cloud data of a foot are used to determine the critical dimensions for making custom footwear. However, automatic and accurate measurement of dimensions, especially girths, is an issue of concern to many designers and footwear developers. Existing methods for measuring girths are primarily based on points or generated triangles, but their accuracy is heavily dependent on the density of the point cloud data. In this paper we present the use of the Radial Basis Function (RBF) surface modelling technique for measuring girths as it has the advantage of being able to operate on unorganised three-dimensional points, so that the generated surface passes through every scanned point, while repairing incomplete meshes. To overcome the high computational expense of the RBF method, local surface recovery, octree division and combination, inverse power method and improved Cholesky factorisation are used. The girth measurements obtained from adopting these approaches are compared against the existing measurement methods. Experimental results demonstrate that the local RBF implicit surface can provide more stable and accurate measurements using relatively less time, proving its value in custom footwear manufacture.

Constrained Surface Recovery Using RBF and Its Efficiency Improvements

RBF based surface rec onstruction from the un-organized three dimensional point cloud provides good implicit representation of 3D object’s shapes, but it has the disadvantage of time consuming. Considering the fact that not the whole surface but only a part of it needs to be recovered in many applications, the method for constrained surface recovery using RBF is studied in this paper. Our approach first t ransforms the partial point cloud with the help of PCA, then generates off-surface points by plane fitting for the local neighborhood of each point. The implicit surface is generated with constructed RBF while the surface edge is constrained by checking whether the marching cube is within the points’ range . To further speed the process of RBF recovery , o ctree based self-adaptive division and combination is used to divide the point cloud into proper groups, inverse power method and improved Cholesky factorization are applied to reduce the time expended in solution s of covariance matrix and linear system. Our method can reconstruct constrained surface in much less computational time, and its efficiency has been tested with the experiments on local visualization and anthropometrical measurement from 3D point clouds .

Accelerated Gaussian mixture model and its application on image segmentation

Gaussian mixture model (GMM) has been widely used for image segmentation in recent years due to its superior adaptability and simplicity of implementation. However, traditional GMM has the disadvantage of high computational complexity. In this paper an accelerated GMM is designed, for which the following approaches are adopted: establish the lookup table for Gaussian probability matrix to avoid the repetitive probability calculations on all pixels, employ the blocking detection method on each block of pixels to further decrease the complexity, change the structure of lookup table from 3D to 1D with more simple data type to reduce the space requirement. The accelerated GMM is applied on image segmentation with the help of OTSU method to decide the threshold value automatically. Our algorithm has been tested through image segmenting of flames and faces from a set of real pictures, and the experimental results prove its efficiency in segmentation precision and computational cost.

Improvements on IPD Algorithm for Triangular Mesh Reconstruction from 3D Point Cloud

A new algorithm is proposed for triangular mesh reconstruction from 3D scattered points based on the existing intrinsic property driven (IPD) method. The improvements include a new approach to determine the seed triangle, a new approach to define the influence region for active edge, and a new approach to select the best active point. The new triangle is also tested with the constraint of geometric integrity. Our algorithm has been tested on some unorganized 3D point clouds. From the experimental results it can be found that our approach has the ability to generate more accurate details in the recovered surfaces.

CUDA Based GPU Programming to Simulate 3D Tissue Deformation

The medical training systems based on virtual simulation are highly desired since minimally invasive surgical techniques have become popular to patients. The training system helps surgeon trainees to acquire, practice and evaluate their surgical skills, and the key component of such a system is to simulate the dynamic procedure such as 3D biological tissue deformation in surgical operation. In our paper, an improved mass-spring model is proposed to represent the biological tissue surface, during which the virtual spring is introduced and utilized to help compensate the weakness of the conventional mass-spring model. Then Verlet integration is adopted to calculate the position of mass points during the deformation process without explicit computation of the velocity values. Finally the bilinear interpolation method is employed to generate one smooth mesh to render the deformed tissue surface. To speed up the simulation performance for surgical tissue deformation, CUDA based GPU computing is adopted, while related data structures and algorithm are designed and implemented for the parallel computation. Our proposed method has been tested by experiments and it has the ability to generate realistic biological tissue deformation images in real time.

Improvements on Electric Field Based Curve Reconstruction from Unorganized Points

Electric field based curve reconstruction algorithm assumed that every point is charged, and the intensity of the electric field reflects distribution and shape of the unorganized points, thus projection of the main ridge of the field surface on 2D plane can be taken as the reconstructed curve. Performance of the method is affected with the appropriate selection of initial points and the speed of their iterative approximations to the limit positions. Our paper makes improvements in two aspects: restrictions are defined for the moving paths of the initial points so that their order is kept and zigzags are avoided; reconstructed 2D curve from the projection of 3D points on one plane are translated as the initial shape to recover the 3D curve. The improved algorithm is applied to generate random number of samples from slices of unorganized 3D point cloud, which can then be easily represented by triangulated or spline surface. Improvements of the algorithm are tested with some 2D and 3D unorganized points, and the results are encouraging.