Bao Sheng Kang’s research while affiliated with Northwest University and other places

In order to improve the robustness of visual tracking in complex environments, a novel multi-feature fusion tracking method based on mean shift and particle filter is proposed. In the proposed method, the color and shape information are adaptively fused to represent the target observation, and incorporating mean shift method into particle filter method. The method can overcome the degeneracy problem of particle. Experimental results demonstrate that this method can improve stability and accuracy of tracking.

The high incidence of accidents and fatalities in the coal mining industry is often attributed to ineffective training methods. To improve the training quality,virtual reality simulation has applied to improve the effectiveness of the training. This paper presents a VR-based system for coal mining safety training. System can provide the intuitive approach for studying of the safety knowledge, and provide the interactive exercises for learning self-rescue and escape in the disasters. An architecture is proposed for customize new training mission quickly. The requirements of functions in the system have been discussed. The implemented system has proved that virtual reality training has the potential to provide more meaningful and effective training to address the identified safety needs in coal mining industry.

In order to improve efficiency of object tracking in occlusion states. A method to detect and automatically track was present in a surveillance system. Firstly, a graph cuts method was employed to segment image from a static scene. To identify foreground objects by positions and sizes of the obtained foreground regions. In addition, the performance to track objects was improved by using the improved overlap tracking method, the tracking method was used to analyze the centroid distance between neighboring objects and help object tracking in occlusion states of merging and splitting. By the experiments of moving object tracking in three video sequences, the experimental results exhibit that the proposed method is better than the traditional method.

In a monocular video scene, in order to improve the efficiency of object tracking and counting under occlusion conditions. The article presents a scheme to automatically track and count people in a surveillance system. First, a modified Gaussian mixture model was employed to determine pedestrian objects from a static scene. To identify foreground objects by positions and sizes of foreground regions which were obtained. Moreover, the performance to track objects was improved by using the modified overlap tracker, the modified overlap tracker was used to analyze the centroid distance between neighboring objects and help object tracking and people counting in occlusion states of merging and splitting. On the experiments of tracking and counting people in three video sequences, the results show that the proposed method can improve the averaged detection ratio about 10% as compared to the conventional work.

A robust statistics approach to curvature estimation on scattered point cloud is presented. The basic idea of this method is fitting a surface to the local shape at a sample point in 3D and the curvatures are computed for this fitted surface. Within a Maximum Kernel Density Estimator framework, the best fitted surface for each point is obtained. Therefore the algorithm is robust with respect to noise and outliers. Experiments show that our method has achieved satisfactory results.

This paper proposes an algorithm to generate and render fireworks in real-time based on programmable graphics hardware, which divides the queue into several regions to save the time when waiting for CPU, and uses the XML external data and XNA framework to achieve the fast simulation of fireworks. The experimental results show that the algorithm can draw the fireworks vividly, meets the demand of real-timing very well, and has a great scalability.