Motion-based object segmentation in video sequences

To make artificial intelligence “see” the world is a primary step to making computers “learn”. Currently, videos are the most common visual resources for robots or artificial intelligence. This research focuses on interpreting the meaningless pixels into semantically meaningful objects. This thesis addresses making computers learn objects from videos automatically based on their motions. Results have shown that this approach can determine the moving objects in short video sequences (2 seconds), no matter how long it has been moving. The learned objects are effective for further recognizing the same objects in different videos. Further research is needed in order to be able to apply the same processes on longer video sequences and scenes that are more complicated.