Using hierarchical temporal memory for vision-based hand shape recognition under large variations in hand's rotation

Abstract

Hierarchical Temporal Memory (HTM), a new computational paradigm based on cortical theory, has been applied to vision-based hand shape recognition under large variations in hand's rotation. HTM's abilities to build invariant object representations and solve ambiguities have been explored and quite promising results have been achieved for the difficult recognition task. The four-component edge orientation histograms calculated from the Canny edge images, have been proposed as the output of the HTM sensors. The two-layer HTM, with 16x16 nodes in the first layer and 8x8 in the second one, has been experimentally selected as the structure giving the best results. The 8 hand shapes, generated for 360 different rotations, have been recognized with efficiency up to 92%. © 2010 Springer-Verlag.