| 研究実績の概要 |
I published the research achievements of this projects in two conference papers. Basically, the following findings are the important points of these papers which demonstrate academic significance:
i. Group recognition: The motion models developed for recognising pedestrian groups are applied to our dataset and it is demonstrated that the group recognition algorithm works with significant accuracy in the presence of strong flow direction, low-to-medium density and for multi-person groups.
ii. Interaction recognition: An interaction recognition stage is integrated into the group recognition method and it is demonstrated that considerable rates of performance are achieved.
iii. Indicators of various gestures: The indicators of coherent motion are investigated with respect to the gestures of interaction and it is demonstrated that different gestures yield different motion patterns.
iv. Variation on motion patterns with respect to interacting peers: The groups are analysed with respect to the interacting side (peer) and it is demonstrated that the peer who is active in interaction is positioned behind within 2-people (2p) groups to have his partner in the field of view. In addition, the not-interacting groups are observed to keep a more disperse distribution compared to interacting groups.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
In this term, I focused on (i) Effect of interaction on group motion and (ii) Attention and state of awareness.
For the first theme, I used an existing extensive dataset of pedestrian trajectories, which is annotated for group and interaction relation. I first confirmed the reliability of the annotations with Cohan's test. Then I proposed a hierarchical strategy for detecting interaction relation. Namely, first the group relation of a pair is investigated. If they are detected to be a group, then their interaction relation is investigated. In this manner, the first stage of the hierarchical strategy (ie group recognition) is shown to perform significantly well. In addition, the combined detection rates of group and interaction relation turn out to be considerably high. In this way, I have completed the initial analysis of the data as in line with the project plan.
As for the second theme of the project, I worked basically on building an experimental setup. I investigated the equipments available in the market and purchased several sensors. Based on the chosen equipment, I prepared the experimental scenario, and recording routines and performed preliminary experiments to confirm the performance. This theme is still under development and interesting results are expected to be achieved after the experiments with actual participants. Therefore, the infrastructure of experimentation has been built as stated in the project plan.
|
| 今後の研究の推進方策 |
A new annotation campaign has recently been finished, which aims at distinguishing the intensity of interaction among the interacting pairs. Two coders annotated the data with respect to the bonding strength between the pairs as well as starting and end times of interaction. In the future, the results of the annotation campaign will be analysed with respect to the following
(i) The annotation of Intensity of interaction is expected to yield distinctions in relation to the bonding strength.
(ii) The temporal variations in the intensity are expected to be resolved using the onset and offset times of interaction.
In addition, the experimental system for measuring attention levels and observing learning strategies is ready to be employed using actual participants. Preliminary experiments have already confirmed that the system works smoothly. The experiments are planned to be carried out after the suspension of project has finished.
|
