language design for activity recognition

29
Language Design Video Activity Recognition

Upload: universite-nice-sophia-antipolis

Post on 14-Apr-2017

188 views

Category:

Engineering


2 download

TRANSCRIPT

Language DesignVideo Activity Recognition

CONTENT

1 - Languages 2 - Activity Recognition3 - Scenarios4 - Experience through Communication5 - Activity Recognition& Scenarios 6 - Designing Scenario Languages/ State of Art 7 - Related Domains8 - Problems9 - Conclusion

It is the method of communication. Either written or spoken, consisting of the use of words/signs/symbols/sounds in a structured way.

WHAT IS A LANGUAGE?

1

According to 20th century Philosophers ,

The roles of ‘grammar’ and ‘syntax’ (rules that govern the structure of sentences) are what makes a language identical.

KEY CHARACTERISTICS OF LANGUAGES

1

It is the method of communication.

in between human and machine

to control or to set the behavior of the machines via instructions.

it has unique set of keywords/symbols and syntax for organizing instructions.

PROGRAMMING LANGUAGE

1

Activities are driven by certain needs where people wish to achieve a certain purpose.

They are usually mediated by one or more instruments or tools.

WHAT IS AN ACTIVITY ?

2

ACTIVITY RECOGNITION

It aims to recognize the activities and goals of one or more agents from a series of observation.

Can be done either by using ‘pre-defined scenarios’ or by ‘machine learning’.

2

2 GENERATION OF ACTIVITY MODELS

3

WHAT IS A SCENARIO?

It is briefly development of events that are projected to occur.

It is consistent and coherent descriptions of alternative hypothetical futures that reflects different perspectives on the past, present and future developments , which can serve as a basis for action.

SCENARIO

3

Human uses language to communicate. [Express, transfer and acquire the information.]

Experience is the result of interaction between a live creature and some aspects of the world which he lives.

EXPERIENCE THROUGH COMMUNICATION 4

EXPERIENCE THROUGH COMMUNICATION

John Dewey states;Human being is the only creature who has the ability to change uncertainties into certainties, ‘the activities’ are the result of that inquiry.

Activities of human beings are continuous with other being’s activities and result of this inquiry.

4

Experiences give birth to Scenarios..

Human beings are born into a certain context. The world merely suggests the objects, it does not give them. Hence the objects are created in the process of changing uncertainty into certainty,

EXPERIENCE THROUGH COMMUNICATION

4

CAPTURING COMPLETE EXPERIENCE / STORY

The information that is transferred via languages are ;

Thoughts, Emotions, Will.

Human perception is capable to capture;

• Thoughts Spoken • Thoughts Written• Thoughts via Sign Language • Emotions via Facial Expressions• Emotions via Body Language

A Suggestion arises,

4

SIGN LANGUAGE

4

SIGN LANGUAGE [Unofficial] :)

4

BODY LANGUAGE

4

Scenarios are the descriptions where details of activities captured fora proper recognition.

By Whom?With Whom ? In Which Context / Sequence / Roles / Duration / Objects ?Where ?When ?and so on…

5

ACTIVITY RECOGNITION AND SCENARIOS

DESIGNING SCENARIO LANGUAGES

6

Scenario Description Language functions can be divided in two category;

1 ) Description of Scenarios

2 ) Recognition of Scenarios

In this study, depending on the purpose we collect scenario description under three headings.

1) Prediction

2) Simulation

3) Test

6STATE OF THE ART

The fact that,

some requires the estimation of best action (move) to obtain a success path (best win scenario).

more intended on future exploration .

useful for recognition performance.

useful for automatic event generation.

So we are not going to concern about prediction requirements of a scenario description language in our study.

6PREDICTION

6

A Suggestion arises,What If we create a directed graph of a certain scenario, in which connected nodes represent next possible action/activity. Then start a random walk on the graph and generate some alternate scenarios, the more visited nodes will have more probable to occur ,So this nodes can be high weighted to obtain a better performance in recognition,In Parallel complement graph can be executed to capture abnormal scenarios, to recognize more early. We can find out causal relations in between activities.

PREDICTION

The fact that,

generally educational function,

simulate system behavior,

information provided by scenario is used to initialize and drive simulation,

strategy games are designed to mimic real life scenarios , matching purpose.

approached to improve autonomous decision making for military purposes.

6SIMULATION

6

VERIFICATION

The fact that,

behavior modeling promotes to create an abstraction of test cases.

helps to ensure verification with a good test coverage.

It is observed that for most of the scenario description languages designed for multi-agent systems, the preferred representation is textual. The reason is that multi-agent system scenarios consists in a set of conversations.

So, future approaches of scenario description languages for multi-agent systems are probably designed with textual representations.

Graphical Representations are easy to learn and apply. It is easy to establish a hierarchical representation and hiding specific details. It is easy to capture big picture.

But for complex systems it is difficult to maintain state compositions between sub actions of coupled main activities. Difficult to customize and revise. Difficult to design.

On the other hand, pedagogically proven fact is that, for a PL acquisition , primarily providing a graphical language then step by step removing graphical representation with a textual language is more suitable.

Concerning graphical and visual languages, for complex systems, the primary concern is to obtain simulation and validation of full system behavior in a complete manner. Hence, languages are far from natural language and not human readable. Also, for a non technical (Computer Science Domain) audience, languages do not have ease of use and are not customizable to obtain generality.

6

OVERALL OUTCOME

There are many domains where scenarios are involved such as;

game, especially strategy games,

military simulations,

crime domain,

film especially animation and science fiction films,

contemporary music composition,

intelligent systems especially multi - agent systems and people detection,

information security , disaster simulation.

7RELATED DOMAINS

Extracting complete vocabulary of sub actions is a problem.

How do we know if we have right scenarios or not. Art of Scenario.

Full communication may need to be captured to obtain a good result for complex scenario/investigation.

There is a casting problem in between actors and their roles. In the current system, object abilities are observable, Actors, model needs to be discussed well such as what can be an actor,

There is always both ways of interaction. There is the passive sentence which guides object point of view, there is active sentence which guides subject/actor point of view. It is questionable how we can model this.

Open question from me , are we really correct to perform, detection of the objects independent from dynamics.

8PROBLEMS IN GENERAL

Languages are the tools that we encode all the information/experience.

Languages can be a strong supporters for us to model dynamic behaviors within its rich , justified , simple and formal properties.

Scenario Description Languages can work for Video Activity Recognition Process by providing variable enough, clear enough, simple enough ,easily prepared scenarios.

CONCLUSION

9

THANKS!