Multi Agent Based Simulation Using Movement Patterns Mined from Video Data

A mechanism for extracting movement patterns from video data with which to drive Multi Agent Based Simulations (MABS) is described. Two types of movement pattern are considered: absolute and relative. The proposed mechanism is fully described in the context of a rodent behaviour MABS. To evaluate the resulting MABS a process is adopted whereby the simulation is “videoed” and the movement pattern generation process repeated (thus completing the cycle). The nature of the simulated movement patterns is then compared with the video data movement patterns. The advantage of relative movement patterns over absolute movement patterns is that they are more generic and this is illustrated in the paper using a case study.