Instrumentation & Measurement Magazine 24-3 - 86

of objects can be categorized into seven main
types according to their
transduction mechanism
namely; resistive, capacitive, piezoelectric, optical
and Organic Field Effect
Transistor-based (OFET)
sensors [7], acoustic, inductive and magnetic
[8]. They produce a 2D
image describing the surface of touched objects.
Other technologies of
tactile sensing relate the
sensors characterizing the
position, acceleration, vibration or force-torque
sensors that sense the amplitude and direction of an
exerted force. Among the
existing technologies, resistive tactile sensors or
Fig. 1. Overall framework for combined use of vision and touch in 3D object representation and recognition.
more specifically Force
Sensing Resistor arrays
of instrumentation techniques are nowadays available for ac- (FSR) are widely used for object recognition and have been
quisition of the geometry of 3D objects; most of them rely on selected in this research [9].
either time-of-flight sensor [3] or triangulation-based systems
for modelling. Radio detection and ranging, Light Detection An Enhanced Computational Model of
and Ranging (LiDAR), shaped light pulse, and sound naviga- Visual Attention
tion and ranging (sonar) are different types of time-of-flight At the heart of the proposed framework for combined use
sensors transmitting and receiving back narrow beam signals of vision and touch for 3D object representation and recogin order to compute the distance between sensor and object nition (Fig. 1) is a computational model of visual attention.
points. They can be used to acquire data on 3D objects through When looking at daily encountered visual scenes, the human
sweeping the beam over the object surface. Triangulation visual system instantaneously processes the huge amount of
based systems basically take advantage of different view- available perceptual information in order to select a subset
points of objects from which specific features are sensed and of relevant and required stimuli. This procedure of selection
matched. Stereo vision, structured light systems and space- or inhibition of perceptual information is referred to as visual
time stereo are examples of triangulation-based systems for attention. Computational models of visual attention, which
3D geometry acquisition. Stereovision is a technique where attempt to mimic the behavior of the human visual attention
two images from different viewpoints are acquired from an system, can be categorized into bottom-up and top-down
object, which are then used to determine correspondences be- models. Top-down attention mechanisms obey different types
tween the images [4]. Since no energy beam is emitted toward of cognitive factors in scene exploration such as, expectations,
the object, stereo vision is referred to as a passive modeling so- or searching for a specific target and prior knowledge, while
lution. In structured-light systems, a light pattern is projected bottom-up models rely on a set of effective features under free
onto the object surface from one viewpoint. An image taken viewing conditions. For the latter, several researches from the
from a different viewpoint can be used to match correspond- field of neuroscience and psychology have identified contribing points and compute the 3D coordinate of surface points [5]. uting features in deployment of visual attention such as color
Spacetime stereo systems project an arbitrarily varying pattern opponency, contrast, curvature, edge, entropy, intensity, orienon the surface of object. Subsequently, they perform feature tation and symmetry. As such, classical computational models
matching using two video streams, taken from different view- of visual attention [10] build upon these features and apply
points over specific space-time windows. Such modeling the center-surround operations exhibited in human receptive
systems are more appropriate for moving or deformable ob- field onto them to produce so-called conspicuity maps. Fig. 2
illustrates an example of conspicuity maps created based on a
jects [6].
In the case of tactile data acquisition, existing tactile center-surround operation on a list of effective features for the
sensors characterizing surface deformations and texture guidance of vision as obtained by an enhanced model of visual
86

IEEE Instrumentation & Measurement Magazine

May 2021

Instrumentation & Measurement Magazine 24-3

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 24-3