Kual-Zheng Lee

Vehicle tracking iterative by Kalman-based constrained multiple-kernel and 3-D model-based localization

In this paper, we propose a novel vehicle tracking system under a surveillance camera. The proposed system tracks vehicles by using constrained multiple-kernel facilitated with Kalman filtering, and then continuously updates the position and the orientation by adopting a systematically built 3-D vehicle model in an evolutionary computing framework. The proposed system can thus successfully track vehicles under occlusion as facilitated by the obtained 3-D geometry of vehicles. Experimental results have shown the favorable performance of the proposed system, which can successfully track vehicles while maintaining the knowledge of 3-D vehicle geometry.

Parallel UPGMA Algorithm on Graphics Processing Units Using CUDA

The construction of phylogenetic trees is important for the computational biology, especially for the development of biological taxonomies. UPGMA is one of the most popular heuristic algorithms for constructing ultrametric trees (UT). Although the UT constructed by the UPGMA often is not a true tree unless the molecular clock assumption holds, the UT is still useful for the clocklike data. However, a fundamental problem with the previous implementations of this method is its limitation to handle large tax a sets within a reasonable time. In this paper, we present GPU-UPGMA which can provide a fast construction of very large datasets for biologists. Experimental results show that GPU-UPGMA obtains about 95 times speedup on NVIDIA Tesla C2050 GPU over the 2.13 GHz CPU implementation.

A Simple Calibration Approach to Single View Height Estimation

This paper presents a simple calibration approach to height estimation based on single view metrology. Instead of calibrating intrinsic or extrinsic parameters, our approach aims to estimate the vanishing points of a stationary camera. The calibration process is formulated as an optimization problem with a novel objective function, in which twelve parameters for estimating vanishing points are defined. The genetic algorithm with Cauchy mutation operator is further used for obtaining robust results. The major advantages of the proposed approach are: 1) it is easy to setup since only a cubic box and some optional line segments are required, and 2) it works without cameras intrinsic parameters. Experimental results show the effectiveness of the proposed approach with digital and analog video cameras.

GPU-UPGMA: high-performance computing for UPGMA algorithm based on graphics processing units

Constructing phylogenetic trees is of priority concern in computational biology, especially for developing biological taxonomies. As a conventional means of constructing phylogenetic trees, unweighted pair group method with arithmetic (UPGMA) is also an extensively adopted heuristic algorithm for constructing ultrametric trees (UT). Although the UT constructed by UPGMA is often not a true tree unless the molecular clock assumption holds, UT is still useful for the clocklike data. Moreover, UT has been successfully adopted in other problems, including orthologous‐domain classification and multiple sequence alignment. However, previous implementations of the UPGMA method have a limited ability to handle large taxa sets efficiently. This work describes a novel graphics processing unit (GPU)‐UPGMA approach, capable of providing rapid construction of extremely large datasets for biologists. Experimental results indicate that the proposed GPU‐UPGMA approach achieves an approximately 95× speedup ratio on NVIDIA Tesla C2050 GPU over the implementation with 2.13 GHz CPU. The developed techniques in GPU‐UPGMA also can be applied to solve the classification problem for large data set with more than tens of thousands items in the future.Copyright

Fast People Counting Using Sampled Motion Statistics

In this paper, we proposed a fast people counting method using the sampled motion statistics. Unlike most traditional approaches uses tracking methods to monitor the moving target until passing through the gate (or region), our method only analyzed motion information in the region of interest to achieve bi-directional people counting. By setting a variety of parameters, a sampled image area according to the pre-defined object size is established in the prior calibration phase. Combining the motion features and its directional status in image sequences, people counts can be estimated through spatial-temporal analysis. Without complex object labeling and tracking procedures, our method shows robust performance in outdoor environments and can be easily implemented on embedded systems. The experimental results show that the proposed method can achieve high accuracy of 94.95% and rapid processing frame rate over 100fps.

Efficient strategy for compressing sparse matrices on Graphics Processing Units

Sparse matrix is used in a large number of important application codes, such as molecular dynamics, finite element methods, path problems, and etc. Much research has proposed several techniques to improve the performance for the sparse matrix operations based on the Graphic Processing Unit (GPU). However, there is no efficient method for compressing sparse matrix on GPU. Hence, in this paper, we design a strategy to efficiently compress sparse matrices based on the concept of GPU. Moreover, we discover the compressing sparse matrix problem that runs on the GPU could encounter some prefix sum problems under the SIMT architecture. We further propose two other types of prefix sum, horizontal prefix sum (HPS) and vertical prefix sum (VPS) in order to solve the compressing sparse matrix problem on GPU.

Efficient image segmentation using a generic and non-parametric approach

Image segmentation is an important process of image analysis. Most of the published approaches for image segmentation need to set appropriate parameter values to cope with the uncertainty problem. However, the parameter values are usually problem dependent and not easily obtained. An efficient image segmentation algorithm using a generic and non-parametric approach is proposed. In the algorithm, the fuzzy-c-means algorithm based on the measurement of the contrast and the compactness between adjacent regions is used to split the image into many small regions first. Then, the validity of existing common edges between regions is checked to determine the mergence probability based on the rank of significance of their contribution using orthogonal array experiments. This algorithm can lead to better computational efficiency and higher segmentation accuracy. Furthermore, the most discriminative regions can be segmented in order with/without a predefined number of regions in obtaining the best and robust segmentation results. Experimental results using artificial and nature images are used to demonstrate the feasibility and efficiency of the proposed algorithm.