| Summary: | 碩士 === 國立中正大學 === 資訊工程研究所 === 103 === Tracking algorithm can be used for pedestrian detection, gesture recognition and aerial object detection…etc. A lot of applications have been found. Our research focuses on multiple target tracking (MTT) algorithm particularly for aerial object detection. We studied three basic MTT：joint probability data association (JPDA), Interacting Multiply Models(IMM) and multiple hypothesis tracker (MHT). Each of them has constraints and will generate errors in real situation. We improved the performance of these algorithms. For example, we combine IMM with JPDA as a new method that has advantages from both sides. We also used clustering and Murty algorithm to greatly reduce the “combination explosion” of joint events. For MHT, we use track-oriented MHT (TOMHT) and greedy randomized adaptive search procedure (GRASP) for optimization MHT.
Since the purpose of our research is aerial object tracking, it is difficult to test with real data. We overcome it by establishing a simulation system, which can randomly generate noise and different targets with different kinds of moving pattern. Finally, we established the method to analyze the characteristics of MTT algorithms, which is another key of our study.
|
|---|
