Xia Limin

Vehicle shape recovery and recognition using generic models

The paper describes a method for vehicle shape recovery and recognition using a generic shape model. The generic shape model, which is able to represent different vehicle classes, is derived by principal component analysis on a set of training shapes recovered automatically from 2D image sequences. With the generic shape model, vehicle recognition can be easily achieved by fitting the model to the vehicle in each image in order to estimate the pose parameters and the shape parameters, which in turn are used to classify the vehicle. Experimental results show the feasibility of our new presented method.

An Image Encryption Approach Using a Shuffling Map

A new image encryption approach is proposed. First, a sort transformation based on nonlinear chaotic algorithm is used to shuffle the positions of image pixels. Then the states of hyper-chaos are used to change the grey values of the shuffled image according to the changed chaotic values of the same position between the above nonlinear chaotic sequence and the sorted chaotic sequence. The experimental results demonstrate that the image encryption scheme based on a shuffling map shows advantages of large key space and high-level security. Compared with some encryption algorithms, the suggested encryption scheme is more secure.

Object tracking using color-based Kalman particle filters

Robust real-time tacking of non-rigid object is a challenging task. Particle filtering has proven very successful for non-linear and non-Gaussian estimation problems. In this paper, a new approach to tracking using color-based particle filers is introduced. The tracked object is characterized by a color probability distribution. The goal of the tracking is to find a B-spline 2D curve in the current image, such that the distribution of the interior region of the curve most closely matches the target model distribution. The Kalman particle algorithm is used to reduce the number of particles needed in tracking mid improve the tracking speed. Results of several experiments are shown to demonstrate the effectiveness of our method.

Wavelet packets analysis of rolling bearing vibration signal and fault testing

Proposes a method for fault testing on a rolling bearing based on wavelet transformation and pattern recognition. Wavelet packets decomposition is used to extract features of dynamic vibration information; K-NN is introduced to test the fault on a rolling bearing. Experiments show that the proposed method attains a satisfactory effect.

Robotic visual servoing based on snakes

In this paper, a robotic visual servoing based on snakes (or active contours) is presented. The approach uses a combination of active contours with image optical flow for tracking image contours related to the object of interest. The active contours provide global information about the position of the object. When visual servoing is combined with the use of active contours, movement of the camera can be controlled by these information. The proposed method fulfils real-time tracking as well as reliability and robustness requirements. Several experiments were performed to verify the potential of our approach.

3D surface reconstruction using fuzzy deformable models

This paper presents a method for 3D surface reconstruction using a deformable model, which integrates fuzzy techniques, deformable model and the level set method. In order to make the result much less sensitive to the initial conditions and noise, fuzzy external forces that drive the model towards surface of object are developed based on a adaptive fuzzy segmentation algorithm. We represent the deformable model as a level set of a 3D scalar function so that the deformable model can reconstruct the 3D surface of the object with complex geometries and topologies. Results of several experiments are shown to demonstrate the effectiveness of our method.

Segmentation of urban traffic scene based on 3D structure

This paper proposes an approach for image segmentation of the urban traffic scene captured from a car-mounted camera. First of all, an improved SIFT feature matching algorithm is adopted for extracting 2D keypoints of the scene. Tracking the 2D keypoints generates the 3D points clouds that can estimate the 3D world structure and motion features. And then Multiple Kernels Support Vector Machines (MKSVM) is employed for sematic segmentation based on motion-derived 3D structure and SIFT features. Experiments show the efficiency and the relevancy of our approach.

Models of reservoir regulation based on RBF neural networks

Reservoir regulation is a very complicated problem with much nonlinear relation. The traditional way of reservoir regulation cannot meet the demands of production. In this paper, a new models of reservoir regulation based on RBF neural network is presented. The historical datum of reservoir regulation is used to train RBF neural network in order to improve the precision of the RBF neural network models of reservoir regulation. The boosting algorithm is used to build an integration-neural network models for reservoir regulation. Experiment results have shown good performance in the actual situation with significant economy benefits.