Ka Keung Lee

Crowd Density Estimation Using Texture Analysis and Learning

This paper presents an automatic method to detect abnormal crowd density by using texture analysis and learning, which is very important for the intelligent surveillance system in public places. By using the perspective projection model, a series of multi-resolution image cells are generated to make better density estimation in the crowded scene. The cell size is normalized to obtain a uniform representation of texture features. In order to diminish the instability of texture feature measurements, a technique of searching the extrema in the Harris-Laplacian space is also applied. The texture feature vectors are extracted from each input image cell and the support vector machine (SVM) method is utilized to solve the regression problem of calculating the crowd density. Finally, based on the estimated density vectors, the SVM method is used again to solve the classification problem of detecting abnormal density distribution. Experiments on real crowd videos show the effectiveness of the proposed system.

Human Driving Behavior Recognition Based on Hidden Markov Models

Automobiles are by now indispensable to our personal lives, but the problem of car thefts threatens the automobile security seriously. In this paper we present an intelligent vehicle security system for handling the vehicle theft problem under the framework of modeling dynamic human behaviors. We propose to recognize the drivers through their driving performances and hope this can help reduce the number of car thefts significantly. Firstly we describe our experimental system-a real time graphic driving simulator-for collecting and modeling human driving behaviors. Using the proposed machine learning method hidden Markov model (HMM), the individual driving behavior model is derived and then we demonstrate the procedure for recognizing different drivers through analyzing the corresponding models. Then we define performance measures for evaluating our resultant learning models using a hidden-Markov-model-(HMM)-based similarity measure, which helps us to derive the similarity of individual behavior and corresponding model. The experimental results of learning algorithms and evaluations are described and finally verify that the proposed method is valid and useful against the vehicle thefts problem.

Real-time estimation of facial expression intensity

Changing facial expressions is a natural and powerful way of conveying personal intention, expressing emotion and regulating interpersonal communication. Automatic estimation of human facial expression intensity is an important step in enhancing the capability of human-robot interfaces. In this research, we have developed a system which can automatically estimate the intensity of facial expression in real-time. Based on isometric feature mapping, the intensity of expression is extracted from training facial transition sequences. Then, intelligent models including cascade neural networks and support vector machines are applied to model the relationship between the trajectories of facial feature points and expression intensity level. We have implemented a vision system which can estimate the expression intensity of happiness, anger and sadness in real-time.

An Intelligent Shoe-Integrated System for Plantar Pressure Measurement

An intelligent shoe-integrated system has been developed to measure both the pressure distribution under eight special plantar regions and the mean plantar pressure during a subjects normal walking. The system mainly consists of 8 force sensing resistors (FSRs) arranged under bony prominences of each foot, a main board based on microprocessor, and a radio frequency (RF) wireless communication module. The digital sampling frequency is 50 Hz which is adequate for the activity of walking. This system is based on support vector machine (SVM) regression for learning the relationship between eight FSR values and the corresponding mean pressure acquired by Pedar insole system (Novel, Munich). Experimental results show that the system can achieve accurate mean pressure estimation with small mean squared error (MSE). Our goal is to provide a reliable and cost-effective system for predicting the value of mean plantar pressure in order to assist patients with musculoskeletal and neurological disorders in the development of normal gait in their daily life.

HUMAN IDENTIFICATION BASED ON GAIT MODELING

Human gait is a dynamic biometrical feature which is complex and difficult to imitate. It is unique and more secure than static features such as passwords, fingerprints and facial features. In this paper, we present intelligent shoes for human identification based on human gait modeling and similarity evaluation with hidden Markov models (HMMs). Firstly we describe the intelligent shoe system for collecting human dynamic gait performance. Using the proposed machine learning method hidden Markov models, an individual wearers gait model is derived and we then demonstrate the procedure for recognizing different wearers by analyzing the corresponding models. Next, we define a hidden-Markov-model-based similarity measure which allows us to evaluate resultant learning models. With the most likely performance criterion, it will help us to derive the similarity of individual behavior and its corresponding model. By utilizing human gait modeling and similarity evaluation based on hidden Markov models, the proposed method has produced satisfactory results for human identification during testing.

Non-holonomic Path Planning of Space Robot Based On Genetic Algorithm

The nonholonomic characteristic of space robot is used to plan the path of the manipulator, by whose motion the base attitude and the manipulator joints attain the desired states. Firstly, the functions of the joint angles are parameterized by sinusoidal functions. Secondly, the objective function is defined according to the accuracy requirement and the constraints of the system state. Finally, genetic algorithm (GA) is used to search the global optimal solution of the parameters. Comparing with other methods, our approach has a number of advantages: 1) The kinematic and dynamic constraints of the manipulator are taken into consideration in the planning process; 2) The dynamic singular point doesnt affect the algorithm since only the direct kinematic equations are utilized; 3) The planned path is very smooth and more applicable in controlling the manipulator; 4) The state converges to the global optimal values. The simulation results verify the method.

Boundary modeling in human walking trajectory analysis for surveillance

Surveillance of public places has become a world-wide concern in recent years. The ability to classify human behaviors in real-time is fundamental to the success of intelligent surveillance systems. The recognition of different human walking trajectory patterns is an important step towards the achievement of this goal. In this research, we utilize the approach of Longest Common Subsequence (LCSS) in determining the similarity between different types of walking trajectories. In order to establish the position and speed boundaries required for the similarity measure, we compare the performance of a number of approaches, including fixed boundary values, variable boundary values, learning boundary by support vector regression, and learning boundary by cascade neural networks. The LCSS similarity approach is also compared with a similarity measure based on hidden Markov model. We found that the boundary establishing method based on learning by support vector regression gives the best results using real-life data during testing.

MULTI-RESOLUTION CROWD DENSITY ESTIMATION BASED ON TEXTURE ANALYSIS AND LEARNING FROM DEMONSTRATION

Crowd density estimation is very important for intelligent surveillance systems in public places. This paper presents an automatic method of estimating crowd density using texture analysis and machine learning. First the crowd scene is modeled as a series of multi-resolution image cells based on perspective projection. The cell size is normalized to obtain a uniform representation of texture features. Then the feature vectors of textures are extracted from each input image cell and the support vector machine (SVM) method is utilized to solve the regression problem for calculating the crowd density. In order to diminish the instability of texture feature measurements, a technique of searching the extrema in the Harris–Laplacian space is applied. Finally, the SVM method is used again to detect some abnormal situations caused by the changes in density distribution. Experiments on real crowd videos show the effectiveness of the proposed system.

An intelligent vehicle security system based on modeling human driving behaviors

This paper presents an intelligent vehicle security system for handling the vehicle thefts problem under the framework of capturing and analyzing dynamic human behaviors. Since human driving skill is a kind of dynamic biometrical feature which is complex and difficult to imitate, it is unique and more secure than static features such as password, fingerprint and face. By utilizing this dynamic property we focus on the research ideal of classifying the drivers into authorized ones and unauthorized ones by modeling their individual driving performance. Firstly, we develop an experimental system architecture. We collect the data of steering, acceleration and braking directly from human driving behaviors as inputs to the system, which aims to achieve better robustness and efficiency. Then, we use fast fourier transform (FFT), principal component analysis (PCA) and independent component analysis (ICA) for data reduction. The features extracted are sent to support vector machine (SVM) for learning and recognition. In the next step, we embed the intelligent classifier into a security system to identify the authorized drivers in response to the real time driving performances. Finally, the experimental results verify that the proposed method is valid and useful against the vehicle thefts problem with a success rate of around 80%.

Context-aware robot service coordination system

We have developed a service-based architecture using Jini to enable the flexible and reconfigurable connection between the interacting components in the distributed robot network. We have extended the original Jini network model in three different ways that allow the system to prioritize tasks using priority queues in the task coordinator service, automatically download proxies to the relevant client based on the conditions of related parties, and automatically deliver service based on contextual information. By using queueing system theory and simulation study, we have characterized the network performance of our architecture and identified areas for optimization