Until now, existing pedestrian detection systemsusually use global features (e.g. appearance or motion) of humanbody to detect pedestrian; however, the detection rate needs to beimproved in many situations since sometimes the global featurescan not be obtained. For example, a pedestrian may be partlycovered by a car or his/her part may hide into the background.Therefore it is essential to adopt some local features of key partsof human body to assist pedestrian detection.In this paper, we propose a method using some key localfeatures of human body to help pedestrian detection. Since theintroduction of additional features will cost the system more time,in order to ensure the detection speed, we firstly use bothappearance and motion global features of human body to selectcandidates, and then use local features of head and leg to dofurther confirmation. In the confirmation stage, we use threekinds of local features (head appearance, face color and haircolor) to detect the head of each candidate; at the same time, wealso choose some particular local appearance features to detectthe leg. The experimental results indicate that this method canimprove detection rate with almost the same detection speed;additionally, it can reduce false alarm sometimes.

In a pedestrian detection system, the most criticalrequirement is to quickly and reliably determine whether acandidate region contains a pedestrian. The detection ability ofwhole system determines directly upon quality of chosen features.However, due to the large number and various types of availablefeatures, it is difficult to find an optimal feature subset andacquire the proper feature proportion at the same time formost traditional methods including AdaBoost Algorithm. Thispaper presents a co-evolutionary method with sub-population sizeadjusting strategy for the feature selection problem in pedestriandetection system. Our method is able to find an optimal featuresubset and adjust feature proportion to a proper state in themean time. Experiments show that our method performs betterthan AdaBoost Algorithm.

In the evolution process of World Wide Web, contentsof web pages play important roles because of their directeffect on linking preference. In this paper, we propose amodel which combines vertex connectivity and content similarityin a proportional manner. Analytical solutions indicatethat our model exhibits a power-law degree distributionwith variable exponent determined by the weight ofcontent similarity. Distribution of content similarity on connectedvertex pairs shows content similar web pages trendto be linked together. Simulation results show our modelyields remarkably agreements of both degree and contentsimilarity distributions with real network.

In World Wide Web, contents of web documents play importantroles in the evolution process because of their effectson linking preference. A majority of topological propertiesare content-related, and among them the clusteringfeatures are sensitive to contents of Web documents. In thispaper, we first observe the impacts of content similarity onweb links by introducing a metric called Linkage Probability.Then we investigate how contents influence the formationmechanism of the most basic cluster, triangle, with ametric named Triangularization Probability. Experimentalresults indicate that content similarity has a positive functionin the process of cluster formation in theWeb. Theoreticalanalysis predicts the contents influence on the clusteringfeatures in the Web very well.

This paper presents a low-cost solution forpedestrian detection using a single optical video camera on amoving vehicle. Since only one optical camera can collect just alittle original information, our system scan two sequential framesto get both appearance and motion information. We use zoomimage and slide window techniques to select the objective region,apply a cascaded classifier combined with statistical learning andSVMto recognize human body, adopt zoom-scale to estimate thedistance from a pedestrian and develop a Distance Transformalgorithm to forecast his/her orientation. This system is suitablefor detecting pedestrians in the range of 0.3-20 meters in the citytraffic with the speed under 50km/h. The test with videos of realcity traffic indicates that our system has got acceptable detectingrate and processing speed.