Predictive Sampler Design for Haptic Signals

Abstract

In this chapter, we study various possible structures of perceptually adaptive sampling strategies for one-dimensional haptic signal. For that purpose, an experimental setup is designed where we record haptic responses extensively for several users. The responses are labeled as perceived (+1) or non-perceived ($-1$). After that, various classifiers are designed to predict the labels of the responses. We have applied several different classifiers based on Weber’s law, level crossing, linear regression, decision tree, and random forest . The classifiers based on the level crossing and the Weber’s law as features have good accuracy (more than 90%) and are only marginally inferior to random forests. The level crossing classifier consistently outperforms the one based on the Weber’s law even though the difference is small. Given their simple parametric form, the level crossing and the Weber’s law based classifiers are shown to be good candidates to be used for adaptive sampling. We have studied their rate–distortion performances and demonstrated that the level crossing sampler is superior. In summary, we have demonstrated that both the level crossing and the Weber classifier-based samplers are good candidates for the perceptually adaptive sampling mechanism for haptic data reduction.