Maximized sensitivity data glove for virtual hand modeling

Maximized sensitivity data glove for virtual hand modeling

1.          INTRODUCTION

This paper is the report of the research project under the module EE7226 Undergraduate Research which conducts on two semester duration research project. Project topic is maximized sensitivity data glove for virtual hand modeling.

In recent years, sensory data gloves have been extensively used along with the popularization of virtual reality (VR) applications. The data glove is a multisensory device that generates a large amount of data and is more complex than other input devices. However, most researchers still adopt the data glove because its natural interfacing characteristic with the human being is the way to improve system robotic control, sign language recognition systems surgery training of medical applications entertainment sports of VR systems industrial manufacturing of CAD/CAM applications.

              A pair of virtual reality gloves is a common device for recognizing hand gestures in virtual environments. Virtual reality gloves can be light and flexible, rendering them comfortable for extended use and are being produced for long enough to offer a stable and quite affordable solution for gesture recognition. There are so many data gloves invented for different purposes. Under mentioned issues of common data gloves designed for specific purposes. That kind of gloves may be used only for a specific task.

1.1.    Application based issues of Data Gloves

Most of the current data glove implementations are designed focusing a specific application. Table 1.1 includes common issues that we can find in such application specific gloves.

Table 1.1 Common application based issues of data gloves

Application	Cause	Issue
Autonomous robotic grasping	Use of  bend sensitive sensors	Failed  to provide enough sensor information in autonomous robotic grasping  in the form of finger and palm coverage
	Sensed Area	Palm has nearly full coverage, but there are still “dead” areas on the fingers
	Wire Count	The current compromise is to bus several sensor pads together to reduce wire count, but it also reduces the spatial resolution of the sensor.
Gesture Recognition	Gesture input	Problem of knowing when gestures start and end
	natural gesturing involves a series of transitions from gesture to gesture	This has a potential drawback of restricting the input rate of a glove
	Speed limitations	Difficult to capture very rapid hand motions
Virtual reality gloves	Gloves which measure bending and abduction of the fingers	No posture data for the whole hand
	Gloves which identify contact between the tips of the fingers	Since no bending data is reported from the gloves, the visual feedback is minimal.

1.2.    Issues of commercially available Data Gloves

Most of the data gloves currently available in the market consists some common issues listed below.

·                  No capability to change the A/D hardware sensor offsets and gains from software permitting the sensors to be tuned to use the full A/D range on a per-user basis

·                  noise in the sensor outputs

·                  sensor support and the quality of  its fit to the user’s hand

·                  need for calibration for new users

·                  High cost

1.3.    Problems Under Exploration

Current data gloves use the currently designed sensors available in the market for specific applications. Hence it’s not possible to sense accurate and efficient hand motion positions for other data glove application. The speed of current data gloves is insufficient to capture very rapid hand motions therefore it has difficult to model a human hand in to a virtual or robot hand. Normal commercial data gloves cannot represent all the DOFs in human hand most of commercial based gloves has lesser DOF compared with human hand such that data glove cannot model in to real human hand. Therefore no posture data for the whole hand so lesser DOF which compared with the real hand. So gloves used for gesture recognition are not accurate enough for fine manipulations or complex gestural recognition. Real time modeling of a hand cannot be done in common data gloves especially with accurate position and orientation of the model.

Also there are no clear defined realistic hand models which cooperate with data gloves. Most of data gloves are made by flex sensors. Flex sensor is a thin sensor which shows variable resistance for bending. Flex sensors are not perfectly suited to measure finger movements because it only identifies the bending not realistic finger movements. Current data gloves focus on robot rather than real Human interaction. When and how objects are grasped, plays a significant role in data gloves. There are low levels of grasping points which can visualize. It is hard to find customizable data glove for virtual hand motion and real world Human interacted application.

2.          HUMAN HAND MOTION ANALYSIS

Due to the issues of the present data glove systems discussed above in order to propose a best solution to address the problems under exploration hand anatomy, movements and motion constraints were analyzed.

2.1.    The Human Hand

Although the hand is a complex biological organ, it can also be considered as a mechanical machine and apply mechanical principles to study it. In this context, it involves three elements: muscles serve as the motor for providing driving force, tendons, bones, and joints transmit the motor’s driving force, and skin and pulp tissues apply the force [1].

We analyzed the various movements of the hand’s different parts and categorized its biomechanical motions.

We can look at hand motions as complex combinations of the movements (rotations around different axes) of different bones at various joints. Fingers have distal interphalangeal (DIP), proximal interphalangeal (PIP), and metacarpophalangeal (MCP) joints, whereas thumbs have interphalangeal (IP), MCP, and carpometacarpal (CMC) joints. Figure 2.1 illustrates the hand’s joints and their degrees of freedom (DOF).

             

Index

Thumb

Middle

Ring

Little

IP

(1 DOF)

MCP

(2 DOFs)

CMC

(2 DOFs)

Wrist

(6 DOFs)

DIP

(1 DOF)

PIP

(1 DOF)

CMC

(2 DOFs)

Figure 2.1 hand’s joints and their degrees of freedom (DOF).

There is only one motion at each finger DIP and PIP joints and at the thumb IP. This is called as flexion bending. The thumb MCP and CMC joints and the finger MCP joint also have side-to-side movement called abduction– adduction. Abduction–adduction is a finger’s motion away from and to the middle finger.

The wrist bones have the most complicated movements, the palm bones’ movements tend to converge to a point in the wrist bones. Thus, for simplicity, we use only one point to represent the wrist joint and its six movements: one for bending (flexion), one for side-to-side movement (abduction– adduction), one for rotation (supination–pronation), and three for displacement in 3D space. Therefore all together there are 27 DOFs in the hand.

2.2.    Hand Motion Constraints

The human hand can’t generate just any random arbitrary gesture. It is constrained. There are limitations on the natural movements of hand parts at their rotation joints [2, 3].

Figure 2.2 and Figure 2.3 illustrates the motion terminology of the hand.

Figure 2.2 Wrist and Finger Movements

Figure 2.3 Thumb movements

3.          PROPOSED SOLUTION

MAGNETIC DATA GLOVE

3.1.    Introduction

Figure 3.1 Coils placement of the hand

Magnetic data glove is based on modern technologies of data gloves. Magnetic tracker which shown in Figure 3.1 has distinct benefits compared to other tracking techniques. There is absence of high precision truly portable human hand and fingertip trackers that are able to provide absolute position and motion data. Data of this format is vital to work in telerobotics, VR and human performance assessment [4, 5, and 6].

Hence the proposed solution is on the development of a new light weight, high precision and dedicated human hand magnetic tracking system. It will be shown that this magnetic tracker has distinct benefits compared to other tracking techniques. The tracker presented is smaller than any other hand tracking system currently reported and moreover provides absolute fingertip tracking in all six degrees of freedom with high accuracy. The section will describe the operation of the device which is based on near magnetic field coupling and consists of one transmitter antenna triad and six receiver antenna triads. The transmitter and receiver antennas triads are enclosed in a cubical casing. The base transmitter station is mounted on the dorsal surface of the hand and the receivers are mounted on the fingertips and any arm point to be tracked.

3.2.    Operation

3 x12mm

Conductors relate to X,Y,Z axis on antenna side

Figure 3.2 Orthogonal coils antenna triad                  Figure 3.3 Constructed antenna                                         

Transmitter antenna triad shown in Figure 3.2 is mounted on the dorsal surface on the hand and receiver antenna triads are mounted on the figure tips and other positions of the hand. When the transmitter antenna is energized by the driving signal given by DSP unit, it produces three orthogonal magnetic fields. Then the receiver antennas (three orthogonal wound coils shown in Figure 3.3) will produce electrical signals according to the strength of magnetic flux they feel, which is produced by the transmitter coils. These produced voltages at the receiver coils also measure at the DSP unit. So this DSP is capable of generating reference signal for the transmitter coils, conditioning the return signal and estimating sensors coordinates in real time. Figure 3.4 shows signals for one sensor coordinates evaluation. This consists of three stage excitation and of capturing induced voltages in the sensor. There is a phase shift between the transmitted reference signal and the return signals. Once nine measured voltages from the sensors and their signs are known the matrix Y representing the induced voltages in the coils can be shown as in Figure 3.5.

Transmitted signal a

Transmitted signal b

Transmitted signal c

Return signal a

Return signal b

Return signal c

ADC sampling

                                          Excitation stage 1                Excitation stage 2                   Excitation stage 3             

Figure 3.4 Signals for one sensor coordinate evaluation

                                                                                   

Figure 3.5 Matrix Y representing the induced voltages

3.3.    System Design

Design of a tracker based on magnetic field coupling is a very delicate task which requires precise components and fast signal processing. PCB design is also critical. The tracker developed in this work consists of single digital signal processor (DSP) and set of additional components chosen with respect to portability and battery power requirements. The heart of the hand tracker is a Digital Signal Processor used for the system. The DSP generates signals, which are after amplification used for driving the orthogonal transmitter coils to excite the magnetic fields. Moreover the DSP is also in conditioning the return signal and estimating the positions and orientations of the sensors. Also the tracker is powered by small size Li-ion battery, which provides energy for long time of continuous operation. Charging of the battery takes about four hours and the tracker can be used during the charging as normal. Captured sensor values are transmitted via Bluetooth which ensures that the tracker can have a cable-free connection to computers, PDAs or other electronic equipments.

3.3.1.                     Design of magnetic tracker

The coils for transmitter and receiver must be orthogonally wound on a nonmagnetic core as shown in Figure 3.3 and 3.4. The coils need to have precise geometry, exactly same value of induction, yet must be as small as possible. Unfortunately, there is no supplier who offers commercially produced coil triads with required features. Therefore we have to prepare these magnetic sensors as our own.

3.3.2.                     Design of DSP unit

The tracker developed in this work consists of single digital signal processor (DSP). We suggest a DSP from Texas Instruments, such as TMS320F2810 clocked at 150MHz [7]. This DSP unit is used to generate the signals and process the receive signals and calculate the orientation and positions of the figure tips. This tracker is small and portable because is generating only weak magnetic field which can be sensed only in short distance (up to 200 mm from transmitter). DSP unit is consisting of DAC module, ADC Module, Signal conditional module and etc. Complete block diagram of DSP unit is shown in Figure 3.6.

Multiplexer

Signal Amplifier

  Filter

Analog to    Digital converter

Signal Conditioning

Orientation and Position calculation

Signal Generation

Digital to Analog Converter

Transmitter

Receiver 1

Receiver 2

Receiver 6

  

              Figure 3.6 Block diagram of DSP unit

3.3.3.                                                                   Signal Generation and DAC module

The magnetic field is excited by a transmitting antenna triad with the drive signal for these transmitter antennas being generated by the DSP. However, since the DSP doesn’t have an integrated DAC alternative options must be pursued including

                  i). an integrated PWM generator, which can produce a sine wave output

                  ii). an external 12-bit DAC.

To ensure higher quality output signals the external 12-bit DAC was selected. A power amplifier follows after the DAC to amplify the signal to create measurable magnetic fields in the field of motion of the fingertips. The power amplifier is a D-class amplifier with minimal temperature looses and a compact package well suited for portable devices. The transmitter’s antenna triad has three orthogonal transmitting coils forming a cube with each side is approximately 12 mm long. Each transmitter coil energize by the drive signal generated by this module.

3.3.4.                                                                   Multiplexer module

Each sensor contains 3 orthogonal coils. For this data glove there will be around five or six sensors used. Therefore we should have 15 to 18 analog to digital converter channels in our DSP module. But the DSP that we suggest here is only have limited Analog to Digital channels. To overcome this issue there should be a multiplexer to interface multiple analog channels to DSP using limited built-in Analog to Digital Converter channels.

3.3.5.                                                                   Signal Amplifier and Filter modules

Since the magnetic field is producing very low induced voltages in the sensor coils it is also necessary to amplify the return signal. A two stage differential amplifier is better to amplify the signals. The differential input amplifier is built from two independent operational amplifiers, which are combined with low-pass and anti-aliasing filters [8]. Differential and common mode filtering is applied at the inputs to the amplifier by a simple filter circuit consisting of passive components.

To boost the signal levels resonant circuits can be used. The resonant circuit radically alters the impedance of the circuit at the tuned frequency, which is good for boosting the signal and moreover for increasing immunity against noise. The signal used in the tracker has a constant frequency around 10KHz.

The offset removal algorithm removes any DC offset component which may be present on the input signal and would disrupt correct operation of the demodulator. It works by maintaining a running average of the input signal and subtracting the result from each input reading. Since the average value of a sine wave is zero over a complete number of cycles, only the DC offset component remains after averaging and this can be subtracted from the ADC reading on a point-by-point basis, effectively AC coupling the input signal.

3.3.6.                                                                   Signal Conditioning Module

Signal conditioning process is associates with smoothing filter, which smoothes the rectified signal magnitude into a smooth and stable digital result. The smoothing is necessary due to the presence of magnetic field distortion in the air surrounding the system. A simple moving average filter was constructed using a window over whole excitation step. The filter runs on a point-by-point basis: each data point from the output of the demodulator is added to a circular buffer, and the average computed by subtracting the oldest value from a running total, adding the newest value, and scaling. The final stage of the signal conditioning consists of measurement sensors geometry correction. Fixed offset and gain adjustments are applied, and the result is used to index into a correction map to retrieve a correction in bits which is added to the reading. Once nine measured voltages from the sensor and their signs are known the matrix Y representing the induced voltages in the coils can be composed.

4.          DETERMINATION OF POSITION AND ORIENTATION

The technique for evaluating position and rotation of the magnetic tracker is described in detail in [9], [10], [11] and involves the following processes.

4.1 Magnetic Field Coupling

The near magnetic field produced by circular loop antenna can be described in terms of radial and tangential components, as shown in Figure 4.1.

        Figure 4.1 Magnetic Dipole Field

The loop shown in Figure 4.1 is excited with a current i(t) = I coswt. The magnetic field produced at a point of distance r and off axis angle d is described completely by radial and tangential components:

              
Ht= M4πρ3 cosδ- (1)

Hρ= M2πρ3cosδ- (2)

Where M = NIA is scaled magnetic moment of the loop, and A and N represents the area and number of turns of the loop [9].

The excitation of a three-axis magnetic dipole source and the resultant three-axis sensor output are conveniently described in vector notation. The excitation of the source is therefore represented by f₁ = [f_1x, f_1y, f_1z] ^T. The number of turns and the area of the three source loops are assumed to be identical. Hence f_1x, f_1y and f_1z represent the amplitudes of the currents exciting the loops of X1 axis, Y1 axis and Z1 axis orientation respectively.

Let the output of a three-axis sensor be similarly represented by f₃ = [f_3x, f_3y, f_3z] ^T, and consider the coupling between that sensor and a similarly aligned source f₂. As depicted in Figure 4.2 each source axis is coupled only to the corresponding sensor axis. Furthermore, since the Y₂, -Y₃ and Z₂, -Z₃ couplings are produced by tangential field components, their amplitudes differ by a factor of -1/2 from the X₂, -X₃ coupling which is produced by a radial field component.

             

                  Figure 4.2 Coupling between aligned source and sensor

The coupling can be described completely in vector-matrix form by

f3=Cρ3S f2= Cρ31000-1/2000-1/2 f2- (3)

Where C = NAG/2π incorporates sensor gain G and the common source factor NA/2π.

4.2 Determining the Position and Orientation

The coupling between a source and sensor of arbitrary position and orientation can be determined by inserting orthogonal rotation matrices [12] into equation (3). These matrices are based upon position azimuth and elevation (α₁ and β₁) and orientation azimuth, elevation, and roll (ψ_1, φ₁, and θ₁) as shown in Table 4.1.

Table 4.1 Orthogonal rotation matrices

	Position	Orientation
Azimuth Rotate X into Y	Tα= cosαsinα0-sinαcosα0001	Tφ= cosφsinφ0-sinφcosφ0001
Elevation Rotate X into –Z	Tβ= cosβ0-sinβ010sinβ0cosβ	Tθ= cosθ0-sinθ010sinθ0cosθ
Roll Rotate Y into Z	Tγ= 1000cosγsinγ0-sinγcosγ	T∅= 1000cos∅sin∅0-sin∅cos∅

Considering the first coupling between the source and a zero-orientation sensor (whose output is f₄) located at (α₁, β_1, ρ) as shown in Figure 4.3. The excitation f₂ of an equivalent source whose X axis is aligned with the line connecting the source and sensor can be determined by rotating the excitation vector of the real source by position azimuth and elevation, thus

f₃ = T_β1 T_α1f₁ – (4)

Figure 4.3 Position and Orientation Coordinates

The coupling to a similarly aligned equivalent sensor f₃ then has the same form as equation (3).

f₃ = (C/ρ³) S f₂

The output of the zero orientation sensor is then found by applying inverse position rotations, thus

f₄ = (C/ρ³) T_-α1 T_-β1 S T_β1 T_α1 f₁ = (C/ρ³) Q f₁ – (5)

The output of the three axis sensor of arbitrary orientation (ψ₁, θ₁, φ₁) is determined by applying orientation azimuth, elevation, and roll rotations to the output of the equivalent zero orientation sensor. Thus

f_{5 = Tφ1 Tθ1 Tψ1 f4 = (C/ρ³) A f4 = (C/ρ³) A Q f1 – (6)}

These expressions will be used subsequently as in [25] to derive the position and orientation finding algorithm.

5.          CONCLUSION

Throughout the report we have discussed how to design a maximized sensitivity data glove for virtual hand modeling. Problems which are discussed at Problem under exploration are solved by proposing a novel idea using a magnetic tracking system embedded in to data glove. Previous researches states that the Magnetic position and orientation tracking is accurate for short distance. Therefore we have proposed a new glove design which uses Magnetic position and orientation tracking for precise measurement of the position and orientation of each finger tip and the palm and which can accurately model a hand with measurements of the glove in virtual reality.

Wireless hand tracker was developed. The tracker has six sensors operating in the range +/-200 mm. A number of tests have been conducted to determine the device accuracy. Results show that static accuracy for position is better than 0.2 mm and orientation is better than 0.1 deg. The tracker reading resolution is 0.05 mm within 100 mm distance form transmitter. Current data gloves use the currently designed sensors available in the market for specific applications. But paper has discussed about a maximized the sensitivity data glove which is not just application specific and can be used whatever the user application is, because we cover all the degrees of freedom of the hand and can represent the entire hand in VR. This tracking system is faster than tacking systems at legacy data gloves.

             

ABBREVIATIONS

Abbreviation	Long Form
AC	Alternative Current
A/D	Analog to Digital
ADC	Analog to Digital Conversion
CAD/CAM	Computer-aided design and Computer-aided manufacturing
CMC	Carp metacarpal joints
DAC	Digital to Analog Conversion
DC	Direct Current
DIP	Distal Interphalangeal
DOF	Degree of Freedom
DSP	Digital Signal Processing
IP	Interphalangeal
MCP	Metacarpophalangeal
PCB	Printed Circuit Board
PIP	Proximal Interphalangeal
PWM	Pulse Width Modulation
VR	Virtual Reality

References

[1]               P.W. Brand, Clinical Mechanics of the Hand, C.V. Mosby, 1985.

[2]               John Lin, Ying Wu, Thomas S. Huang, Modeling the Constraints of Human HandMotion, Beckman Institute, University of Illinois at Urbana-Champaign

[3]               Beifang Yi, Frederick C. Harris Jr., Ling Wang, and Yusong Yan, Real time Natural Hand Gestures, IEEE Computer Society, 2005

[4]               Bezdicek, M., "Precise human hand tracking and motion mapping onto robot using UI methods", PhD thesis draft, 2006

[5]               Corke, P.I., "A Robotics Toolbox for Matlab", IEEE Robotics and Automation

              Magazine, pp.24-32, also at www, 1996

[6]               Craig, J., "Introduction to Robotics Mechanics and Control", Addison-Wesley 1989,ISBN: 0201095289

[7]               Ascension Technology Corporation, http://www.ascension-tech.com, internet page,2006

[8]              http://www.maxim-ic.com/app-notes/index.mvp/id/928

[9]              Frederick H.Raab, Ernest B.Blood,Terry O.Steiner, Herbet R. Jones, Magnetic      Position and Orientation Tracking System, IEEE Transactions on Aerospace and Electronics Systems vol AES, No 05,September 1979

[10]               Raab, F.H., Burlington, V., "Remote object position and orientation locater", US patent 4314251, 1982

[11]               Eugene Paperno, Ichiro Sasda, and Eduard Leonovich, A New Method for Magnetic Position and Orientation Tracking, IEEE Transactions on Magnetics, Vol 37, No 04,July 2001

[12]              http:// euclideanspace.com/maths/algebra/matrix/orthogonal/rotation/index.htm

Department of Electrical and Information EngineeringPage 3