추정 정밀도 향상을 위한 다중모델 및 스무딩 기반의 강인한 추종기법에 관한 연구

Abstract

In past, automotive technologies focused on mechanical operation and performance for a long time. But development of computing performance has led to large amount of data processing possible. This made many control systems developed using electric devices. There is now a transition of electric car in the automotive industry so that electric devices are used for driver's convenient. The technologies contribute to driver’s convenient when they drive the car. However, not all the technology for intelligent vehicle has a good effect because some of options directly connect to human's life. Therefore development of the vehicle technology has very conservative side and takes only the proved technology for a long time. Some technologies in automotive industries have been taken from the military technologies or the aircraft technologies like radar cruise control. Among those technologies, the moving object tracking methods have been studied for a variety of purposes. For example, sensors attached to car detect distance of objects forward for a safety system. In this thesis paper, the proposed algorithm focuses on a moving object tracking using sensors attached to a vehicle or a mobile robot in practical applications. The target could be vehicles or obstacles. We need to figure out whether it is target or not. Therefore, raw data from laser scanner should be clustered and find center point of object for the initial tracking. Next, we implement a filtering algorithm for the first time in order to get the initial states of object. After this procedure, if the detected object is target, we start tracking the target with tracking algorithm. In order to enhance the state accuracy, this paper presents an obstacle tracking method taking advantage of strength among different kinds of filtering methods. The point of this paper is that the filtering accuracy is enhanced by the initial tracking and the state-augmented method, so called smoothing, reducing peak error during transition mode.