This paper describes the development of algorithms for the recognition of (Army) Dismounted Infantry signal gestures. The set consists of seven static poses (two arms), and ten dynamic (one arm and two arm) coordinated movements. The tracker data was generated by 6-DOF tracking devices attached to the wrists and back of the subject. The approach for static gestures is based on template matching. For the dynamic gestures, two approaches were used. One based on tracing 3d points over time, using a state machine like algorithm and the other using analysis of the trajectories of the sensors. The development of the recognition system is not yet complete, but partial results are reported.

A set of .tif files are on the ftp server, under pub/vrtp/Jove/Gestures which show the specific gesture set.

Team members :

Dan Searles - Algorithm development/implementation. Gregory Baratoff - Algorithm development/implementation. Jerry Smith - Background information/analysis. Brian Bohmueller - Report generation/editing.