Development of Registration and Manipulation Techniques for Robot-Assisted Minimally Invasive Brain Surgery

The goal of object tracking is to establish a stable track for objects of interest in successive frames. It has been an essential technology for human survival and has significant potential for human progress. For example, the technique can be used for airspace surveillance, military tracking of targets, submarine and whale tracking, autonomous robot navigation, intelligent traffic analysis, and intelligent video surveillance. They are also used in the study of cell biology to study cell fate under different chemical and environmental influences. In tracking, there are several practical problems that can render failed targeting, such as morphological variation due to the motion of objects, change in shadowing and lighting condition, and blocking of the target. The purpose of this project is to develop a dynamic optical tracking system based a PTZ camera platform to stably track any selected object and keep the targeted object image in the center of video images. This successive proposal follows an on-going one-year project. In the project, we have already developed a working optical tracking system which includes the provision of user interface for the assignment of targets, and integration of tracking algorithm with the motion control of a PTZ platform. Based on a well-known TLD (Tracking Learning Detection) tracking algorithm, we have parallelized the most computationally intensive detection part of the algorithm. Besides, we have implemented an update mechanism in the Compressive Tracking part, together with Adaboost meta-algorithm to improve correctness of the classifier. Stable and reliable tracking heavily relies on learning capability to adapt for variations in target images. In this new project, we will research into practical issues of morphological variation, change in shadowing and lighting condition, and obscure effects, focusing on the difficulty to trace back a lost target. We will implement compressing techniques to accelerate execution, use Halton sequence to decide the comparison regions, and study their effects on correctness in real-time tracking. Learning mechanisms with multiple templates for potential target comparison will also be developed to cope with the situations of dramatic pattern variations to enhance tracking correctness.

Project ID:PB10408-5700
External Project ID:MOST104-2221-E182-008-MY2