Image Feature Detection And Application For Two Targets

Object detection, recognition and tracking are very important contents for Pattern Recognition and Intelligent System, and it is also a main part of Machine Vision. In complex natural scene, using information process technology of computer to analyze scene, to detect, recognize, or track the objects that we are interested in, one can obtain their qualitative and quantitative features so as to descript the image precisely, overall, reliably and construct varies different and complex application system. That is basis of the automation and intelligent and also is a difficult and emergent task. On the concern of this and on the benefit of sustentation fund of the Nature Science and the sponsor of Huawei Company, we select two typical objects as our investigation subjects: near view human face and distant frontal looking of airport runway. We have found out some new techniques of near view human face detection, techniques of detecting and tracking of the distant frontal looking airport runway. And we have also found some new techniques for their corresponding applications. Following contents is our main work of this dissertation.First, we differentiate some concepts of special targets and make the concept of these inkling targets'clearly. A survey has been made in detecting, tracking and application of human face and the distant frontal looking airport runway. In the survey, the actuality state of these two subjects are discussed, include their significance and their foreground.Next, without translation, rotation and scaling transform, we propose an attention-drive approach to estimate the interesting human face region directly guided by our novel Arrow Guide Strategy (AGS) algorithm. Therefore a dramatic speedup is achieved and it is efficient and robust. And so we could use it to segment video scene into foreground and background. If pay more on foreground according to the human vision characteristic and make the bit allocation between foreground and background suitable, we could improve the subjective feeling of video image. It is prominent especially in low bit rate. In this dissertation, we put foreword a method of bit allocation among foreground and background and a strategy that the foreground has priority of encoding, as well as a method of quantification of optimized sequence, all this are based on H263/TMN8 and preamble human face estimation algorithm AGS. Next, we have studied how to locate the human face organs accurately in near view, especially to the apple of human eyes. At first, a novel gradient of rectangle ring has been put foreword to extract human eye candidates in image. Then we use a discrete Hidden Markov Model (HMM) that included explicit state duration density to screen the human eye candidates. At last, a model of sparse retinal sampling grids and its transformed matrix, which is used to analyze the texture feature of human eye by Gabor filters, are presented. All these process guarantee a high precision to the localization of the human eye apples. The test result shows that these algorithms have lower localization errors than Jesorsky's.Another main subject of our dissertation is Runway Detecting and Tracking of an Unmanned Aerial Landing Vehicle, and its application. In our paper, we use pin-hole perspective projection model, the constraint condition of the rectangle in inertial space, the front shot constraint condition of the target, as well as the clustering algorithm to identified the runway and output its position and orientation in image space. After that, we extend the basic runway-detection algorithm to the runway tracking. A simplified decoupling control system of UAV via the input-output feedback technique is presented. And a full filtering strategy using particle filter can guard against potentially catastrophic results and improve the detection rate. The whole algorithm of our paper can be treated as a special vision sensor for landing equipment of UAV.To the application of the runway detection and tracking, a theoretical study of stereo vision system for an unmanned aerial vehicle to land autonomously is given by this paper. And the feasibility of combining two cameras with one gyro meter to measure and track the position and attitude of the vehicle without the aid of Global Positioning System (GPS) is also studied. The measurement error is highly cared for by the general measuring and tracking models.In conclusion, all the new methods, new algorithms or techniques in our paper have been tested and proved to be efficient. Analogy exists between objects in some sense, so what we have done is very important for machine vision and it is a basis work for some important applications of intelligent machine.