Tuesday, November 15, 2005

[Vision] Walking Human Avoidance and Detection from A Mobile Robot using 3D Depth Flow

This paper shows walking human avoidance and detection behavior of a mobile robot using 3D DepthFlow. we propose 3D Depth Flow, that is able tomeasure 3D motion vector of every pixels betweentwo time sequential images. First, a denition of 3DDepth Flow and a simple 3D Depth Flow calculationmethod are denoted. Then an implementation of real-time 3D Depth Flow Generation system using stan-dard PC, and experimental results are denoted. Fi-nally, as an application, walking human detection andavoidance task using a mobile robot in real environ-ments is shown.
Paper is available at
Keyword: perception, humanoid, IEEE (GOOGLE)

1 comment:

Burt Ho said...

This paper describe a method of combining stero vision and potential energy with gradient approach to avoid walking people. But my question is (1) How they solve the correspondence problem between two CCD image. (2) Is this method fast enough? (3) They using PC at mainframe and energy cost?
I think some solutions can fix their algorithm (1) Using range scanner to avoid correspondence problem (2) Using Wi-Fi for computing at PC, robot only receive and execute command for power-saving.