Realization of real-time vision-based control of an inverted pendulum using HOG features and SVM classifiers (Master Thesis)
For the reasons of less expensive commercial cameras and develop ed image processing algorithms, the vision based control is becoming more and more popular and important. However, how to enhance the real-time capability and robustness in vision based control systems is always the key p oint of the leading-edge researches. This thesis aims at improving the vision based inverted pendulum real-time control system in two aspects, application of robustness image processing algorithm, i.e., Histogram of Oriented Gradients (HOG) with Linear Support Vector Machine (SVM) classifier, and Kalman prediction based control schemes with online disturbance estimator. In the system control loop, linear-quadratic-Gaussian (LQG) method is implemented to design an optimal controller. In this vision based inverted pendulum system, a CCD web cam observes the pendulum plane as a vision sensor. The captured real-time images are transmitted from the camera to PC. The PC works not only as a system controller to generate the required actuation signal, but also as a part of vision sensor to analyze the received image. By using image processing methods on PC, the current cart and pendulum positions can be obtained from the real-time image. After receiving control signal, DC-motor drives the cart to ensure the pendulum maintain at the equilibrium point. All the proposed algorithms and control methods are verified successfully in the practical experiments.
Name: Mr. Heng Li
Universität Duisburg - Essen (Campus Duisburg)
Automatisierungstechnik und komplexe Systeme
Prof. Dr.-Ing. Steven X. Ding
Bismarckstraße 81, Gebäude(BB)
Eingang: Oststraße 99