Discussion

On this page we want to open the discussion to further elaborate and refine the original conceptual framework and classification of movement patterns.

Please feel free to comment.

Comments and Questions from a QSR and GIS Perspective

The following comments and questions are given by a seminar group of S. Hazarika, F. Dylla and several Ph.D. students at the University of Bremen with a background on QSR (Qualitative Spatial Reasoning), GIS, and Reasoning about Action and Change.

* Data-Driven Perspective

The approach in general is inspired by data-driven approaches. A qualitative perspective is not reflected in the current approach.

* Problems in the definitions of movement patterns (Table 1):

• the denomination 'primary and secondary derivatives' is strange from a physicians perspective as position, speed, and acceleration are dependant by derivation over time
• in this classification only an absolute frame of reference (FoR) is reflected, but e.g. in modelling or recognizing navigation behavior the intrinsic orientation of objects are necessary (i.e. a relative FoR). From my point of view such an orientation should be regarded as primitive as well. Alternatively, one could use the primitive 'pose' (x,y,\theta) instead of 'position'
• Although the focus of the paper is to deal with MPOs regions should be regarded as spatial primitive as well. Nevertheless, it is not straightforward how to derive a 'spatial extent' from a set of positions. I would rather expect a distance or a set of distances. Additionally, you restrict to what we call continuous motion (also confer to the discussion of dis-continuity in path type), but not to the creation/birth or disappearance/death of objects or regions. For example, a water hole that disappears in summer or a flock that splits up.
• Why is there no spatio-temporal primitive? From our QSR (Qualitative Spatial Reasoning) perspective the primitive seems to be a snapshot (x,y,\theta,t)

* Number of Moving Objects involved

In QSR we deal with relations of arity n. In your classification it is assumed that only binary relations are possible, but what about n-ary relations with n>2? For example, if a flock must contain at least three MPOs if there are N=n MPOs. Therefore, a classification from a QSR (relation) perspective should contain the categories: N<n, N=n, and N>n.

* Path type

We (QSR, Reasoning about Action and Change) have a different (or more diverse) understanding of dis-continuity. A car moving around in a city definitely performs a continuous motion in space and time on a physical level as it cannot vanish at one point and reappear at another one. On a conceptual level this might happen when the car goes thru a tunnel. One could see the discontinuity you are describing is in the domain of speed. The classification you are giving here is already quite application driven and in that sense no objective classification. We would prefer to describe path types more in terms like: straight, un/stable, smooth, uniform, etc. (which are no objective parameters as well)

* Influencing factors

On the first level we would only distinguish two different groups: 1) the intrinsic agent properties and 2) the environment as spatial constraints and other agents are part of the environment. On the second level we had a (not completed) discussion on two classifications of environment on totally different levels:

    1) spatial constraints, other agents, other non-spatial constraints (e.g. wind). We note, that we are aware that any entity in an environment could be modelled as an agent.

    2) spatial, temporal, and other constraints

* Scale/Granularity

A good starting point for further discussions on granularity and scale is work by Stell or Stell/Worboys as well as work by Hornsby/Egenhofer.

* Classification of Patterns (Fig. 1)

When discussing Fig.1 it turned out that the terms "x patterns" where ambiguous or misleading at some point for us ... we know that we are in the territory of 'pattern recognition' here.

We classified 'generic patterns' as '(objective) observations' whereas 'behavioral patterns' are 'interpretations' containing additional domain dependant knowledge.

The term 'primitive pattern' was misleading as well as we don't understand them as patterns rather than properties of patterns or 'pattern primitives' from which the patterns themeselves can be build.

Other terms mentioned which might be not misleading were: pattern properties, parameter, archetype, or prototype.

Additional interesting literature in the QSR community:

• Introduction to QSR:
  • Anthony G. Cohn, Shyamanta M. Hazarika: Qualitative Spatial Representation and Reasoning: An Overview
  • A. G. Cohn and J. Renz, Qualitative Spatial Representation and Reasoning,
• Regarding different topics mentioned above:
  • Zhan Cui, Anthony G. Cohn, David A. Randell: Qualitative and Topological Relationships in Spatial Databases.
  • Baher A. El-Geresy, Alia I. Abdelmoty: Qualitative Representations in Large Spatial Databases
  • Kathleen Hornsby, Max J. Egenhofer: Modeling Moving Objects over Multiple Granularities
  • John G. Stell: Granularity in Change over Time