The XML translator is a java applet that reads a XML file and generates the equivalent LISP codes. The input file is written in XML (Extensible Markup Language) and describes a story. The output is a story file in LISP and can be understood by Meta-AQUA. This file can then be input into Meta-AQUA for further processing.
The demo shows a web page with 3 frames. The frame on the left displays 3 images that can be used to represent a story. Clicking on each image will display the XML codes associated with the image in the upper frame of the browser. Next to each image is a link labeled "LISp". Clicking on this link initiates an applet. This applet takes the XML codes and translates them into the equivalent LISP codes. The generated LISP codes are displayed in the lower frame of the browser.
The first picture shows a man trying to unlock the car not with a key but with some stick. The incident is about a potential car theft which is being reported to the police by a witness. Clicking on the picture displays the XML codes for this incident on the upper frame. Clicking on the LISp link displays the story file for this incident in the lower frame. The story indicates an anomaly of car-theft versus house burglary.
The second picture shows a policeman. This story is about the police upon investigation let the man go. This story shows an anomaly of letting a suspect go instead of putting him to jail.
The third picture shows the man driving away in the car. The story describes the explanation for the above two anomalies. The man lost his car key and thus was trying to unlock the car by some other means. The combined story file can then be input to Meta-AQUA for introspective explanation.
Much studies have been done on image understanding where the objects are visualized, understood and conclusion drawn. Video understanding carries the study a step further. Given a video which comprises of multiple images, a series of primitives are derived that describes the contents and relationships among different frames. Such primitives can be in the form of keywords or frames types that can be handled (or defined in ) by Meta-AQUA. This research develop methods to understand these primitives, derive video events, infer goals of characters, summarize the video contents and eventually predict the video outcome. Such study will use case-based reasoning and machine learning technique. The initial phase of this research is to derive an indexing technique for the cased-based library which can store the primitive sets and their related GOAL description. With the completion of the index schema and fundamental framework, it should be simple to derive the video contents using CBR technique. The result of this research can be applied to facilitate web-based search of video and automated security surveillance.
The Meta-AQUA system accepts only input file written in LISP. To make it easier to work with web-based application, the translator parses a story file written in XML into a DOM tree structure and generates the LISP codes for the story. This generated code can then fed into Meta-AQUA for further processing. This demo is a proof-of-concept prototype and can be considered as an secondary part of the thesis.
The plan for this research is as follows:
- Front end XML -to -Lisp translator. (Proof Of concept - see demo)
- Derive a knowledge weak index primitive and knowledge strong ( domain specific) indexing scheme. The law enforcement field is initially selected as the experimental domain.
- Investigate and enhance indexing scheme for the case library.
- Develop a software framework for the case library.
- Develop goal-based framework for machine learning technique.
- Develop process flow to integrate CBR and learning technique for primitive understanding.
- Integrate above steps to arrive at a content summary and prediction.
- Present research findings.
If you need more information, please contact me
or my thesis advisor