VIDEO TRACKING USING SUPPORT VECTOR MACHINE (SVM) FOR VISUAL SERVOING APPLICATION

• Main Author: Oliver Asali, Modestus
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/35663
• Institution: Institut Teknologi Bandung

Visual servoing has been widely used in various sectors such as manufacturing, transportation, health, military, and security. In the security sector such as the use of video surveillance, visual servoing requires the ability to be able to track any arbitrary object specified at the start of tracking process on various forms of objects in complex environments and in various lighting conditions. The classical approach in feature extraction, such as using fiducial marker, is not possible in this condition. Instead, the integration of video tracking algorithms can be done to handle the feature extraction process. Video tracking is the process to predict the position of an object’s feature(s) in a sequence of images recorded by the camera continuously. In order to be used in real-time systems such as visual servoing, video tracking must be efficient, high precision, and robust. In addition, visual servoing applications are also required to have the ability to adapt to changes in their environmental situation. For this purpose, we use structured-output tracking with kernels (STRUCK) algorithm. This algorithm is one of the video tracking algorithms that has the ability to learn to adapt to changes in the environment (machine learning) using support vector machine with a relatively low computational load while still maintaining its accuracy. Performance test of tracking algorithm is performed for various conditions such as rotation, scale, illumination changes, occlusion, and also clutter with stable accuracy results. Testing the results of integration with the visual servoing system also shows good performance and can be used for tracking various objects in both indoor and outdoor environments.