Peijun Ma

Rule learning for classification based on neighborhood covering reduction

Rough set theory has been extensively discussed in the domain of machine learning and data mining. Pawlaks rough set theory offers a formal theoretical framework for attribute reduction and rule learning from nominal data. However, this model is not applicable to numerical data, which widely exist in real-world applications. In this work, we extend this framework to numerical feature spaces by replacing partition of universe with neighborhood covering and derive a neighborhood covering reduction based approach to extracting rules from numerical data. We first analyze the definition of covering reduction and point out its advantages and disadvantages. Then we introduce the definition of relative covering reduction and develop an algorithm to compute it. Given a feature space, we compute the neighborhood of each sample and form a neighborhood covering of the universe, and then employ the algorithm of relative covering reduction to the neighborhood covering, thus derive a minimal covering rule set. Some numerical experiments are presented to show the effectiveness of the proposed technique.

Kernelized Fuzzy Rough Sets Based Yawn Detection for Driver Fatigue Monitoring

Driver fatigue detection based on computer vision is considered as one of the most hopeful applications of image recognition technology. The key issue is to extract and select useful features from the driver images. In this work, we use the properties of image sequences to describe states of drivers. In addition, we introduce a kernelized fuzzy rough sets based technique to evaluate quality of candidate features and select the useful subset. Fuzzy rough sets are widely discussed in dealing with uncertainty in data analysis. We construct an algorithm for feature evaluation and selection based on fuzzy rough set model. Two classification algorithms are introduced to validate the selected features. The experimental results show the effectiveness of the proposed techniques.

Online UAV path planning in uncertain and hostile environments

Taking uncertainties of threats and vehicles’ motions and observations into account, the challenge we have to face is how to plan a safe path online in uncertain and dynamic environments. We construct the static threat (ST) model based on an intuitionistic fuzzy set (A-IFS) to deal with the uncertainty of a environmental threat. The problem of avoiding a dynamic threat (DT) is formulated as a pursuit-evasion game. A reachability set (RS) estimator of an uncertain DT is constructed by combining the motion prediction with a RRT-based method. An online path planning framework is proposed by integrating a sub goal selector, a sub tasks allocator and a local path planner. The selector and allocator are presented to accelerate the path searching process. Dynamic domain rapidly-exploring random tree (DDRRT) is combined with the linear quadratic Gaussian motion planning (LQG-MP) method when searching local paths under threats and uncertainties. The path that has been searched is further improved by using a safety adjustment method and the RRT* method in the planning system. The results of Mont Carlo simulations indicate that the proposed algorithm behaves well in planning safe paths online in uncertain and hostile environments.

Comparative analysis on margin based feature selection algorithms

Feature evaluation and selection is an important preprocessing step in classification and regression learning. As large quantity of irrelevant information is gathered, selecting the most informative features may help users to understand the task, and enhance the performance of the models. Margin has been widely accepted and used in evaluating feature quality these years. A collection of feature selection algorithms were developed using margin based loss functions and various search strategies. However, there is no comparative research conducted to study the effectiveness of these algorithms. In this work, we compare 14 margin based feature selections from the viewpoints of reduction capability, classification performance of reduced data and robustness, where four margin based loss functions and three search strategies are considered. Moreover, we also compare these techniques with two well-known margin based feature selection algorithms ReliefF and Simba. The derived conclusions give some guidelines for selecting features in practical applications.

Fast Pedestrian Detection Using Slice-Based Motion Analysis

This paper presents a novel slice-based approach to detect pedestrians in still images. A pedestrian is divided into limited numbers of slice-based sub-regions through a spatio-temporal slice processing. First, sub-regions of interest are detected in different spatio-temporal slice images. Then, a clustering algorithm is proposed to combine these sub-regions into individual pedestrians based on their consistent motion patterns. Experimental results show that the proposed method can tolerate sub-region missing, i.e. partial occlusions, and reduce the computational cost.

Image Secret Sharing and Hiding with Authentication

Recently, Lin-Tsai, Yang et al., and Chang et al. proposed image secret sharing and hiding schemes with authentication. The secret image is shared and hided into ordinary cover images to form the stego images so as to be transmitted securely. Unfortunately, there is a common weakness that each stego image should be expanded to 4 times of the secret image. In this paper, we propose an enhanced (t, n) threshold scheme with smaller size expansion of the stego images. The size of each stego image is reduced to 3.5/t times of the secret image with the image quality better than the previous schemes. In addition, the authentication property is improved based on hash function. Finally, the experimental results show that our scheme is superior to the compared schemes.

Driver status recognition by neighborhood covering rules

Driver fatigue recognition based on computer vision is considered as a challenging issue. Though human face carries most information related to human status, the information is redundant and overlapped. In this work, we concentrate on several fatigue indicating areas and three types of features are extracted from them. Then a neighborhood rough set technique is introduced to evaluate quality of candidate features and select the effective subset. A rule learning classifier based on neighborhood covering reduction is employed for the classification task. Compared with classic classifiers, the designed recognition system performs well. The experiments are presented to show the effectiveness of the proposed technique.

Combine Feature Selection with Timing Sequence Energy Analysis for Driving Drowsiness Detection

In this work, we try another way by introducing a novel method which combines feature selection with time sequence analysis techniques to estimate driving drowsiness. Kernelized fuzzy rough sets based technique is used to evaluate quality of candidate features and select the most useful one. S transform is adopted for blink energy analysis. Finally the experiments on three blink sequences with dissimilar fatigue degree are used to validate our ideas.

Multitarget PHD Particle Filter Tracker Based on Single-Target PHD

Probability hypothesis density (PHD) filter is a new practical method for tracking multiple targets. However, to obtain the output of the PHD filter, peaks-extraction and association between frames is needed to estimate the target states and the individual target trajectories. A new PHD filter tracker based on single-target PHD is proposed. The method estimates target states by decomposing PHD into single-target PHDs, and associates target location estimates between time frames based on the additional labels of particles. Simulation results demonstrate that the new algorithm provides more accurate trajectory estimations and is more efficient than the PHD tracker with k-means algorithm and the usual particle-labeling association.