Browsing by Author "Zahedi, Edmond"

Now showing 1 - 3 of 3
  • Thumbnail Image
    ItemOpen Access
    A New Approach To Amputee Training Using Computer Graphics
    (1995) Zahedi, Edmond
    In this paper, a new approach is presented for the training of amputees for an effective fitting of upper-extremity myo-electric prostheses, Electromyogram (EMG) signals from the remaining muscles of the stump are amplified and filtered, digitized with 8 bit resolution and fed htro a microcomputer with a sampling rate of 2500 Hz, The extracted feature is the integral absolute value of the biceps and triceps. A fuzzy classifier is used for clustering and classification. The EMG signals are processed and the decision of the classifier animates a graphically simulated 3 degree of freedom prosthesis on the microcomputer monitor. Five healthy persons have been trained with this system It took less than 10 minutes for the subjects to familiarize with the operation of the system. The error rate was less than 5%. The main advantages of such an approach are thought to be: training of amputees before using a real prosthesis by remembering them the concept of "muscle state", easy evaluation of the misclassification error rate of different algorithms, expert man-power time saving, accurate follow-up of the amputee, more availability of the training set and self-paced learning so less frustration of the amputee.
  • Thumbnail Image
    ItemOpen Access
    A New Concept In The Evaluation Of Cybernetic Actuators Control Using Virtual Reality
    (1995) Zahedi, Edmond; Miyake, Hitoshi
    In order to control the movement of a cybernetic actuator the EMG signal is generally used as a source of command. This signal has to be processed in order to extract relevant features, which are then classified. Many schemes exist today in both feature extraction and classification, each one claiming to reduce the error rate and there has been some approaches in order to assess the input-output characteristics of prostheses. This paper vvill introduce a new concept in developing a unique platform using virtual reality (VR) tools for evaluating both the different schemes of EMG signal processing and cybernetic control. The design follows a modular approach allowing for the change of each module (analog signal conditiorming, data acquisition, signal processing, actuator control, VR aspects) accordingly to the specific needs of an application. The foreseen applications of this work are performance evaluation of EMG signal processing algorithms for prosthesis control in real conditions, performance evaluation of the motor control schemes by executing real tasks, selection of the optimum scheme for a particular application (spatial, medical surgery, underwater, etc...) and training of amputees or future users of the system in real conditions where "real conditions" means the VR simulated environment.
  • Thumbnail Image
    ItemOpen Access
    Low-Cost, Real-Time Emg Signal Processing For Prosthesis Control Using Dynamic Patterns
    (1995) Danesh Pajoug, Hamid; Zahedi, Edmond
    In this paper an approach to control a 3 degrees of freedom (DOF) below-elbow prosthesis is presented . Myoelectic patterns generated by flexor and extensor muscles during the initial phase of muscle contraction are used. It has been shown by previous researcher s that specific "dynamic" patterns exist during this period. The aim of this work was mainly to achieve high classification success rate with a minimum mathematical complexity in order to reach a low-cost upper-limb pr osthesis design. A special purpose hardware based on an Intel 87C52 microcontroller has been developed. EMG signals are amplified by an instrumentation amplifier and band-pass filtered before being digitized by an A/D converter. A serial link to a PC allows (in an off-line mode) the setting of different parameter s such as sampling rate and feature to be extracted. The mean absolute value (MAV) of the EMG computed over windows consisting of 5 samples was used as a feature. Only the first 160 ms of the EMG was used because during this period obtained patterns show good separability. A new motion was detected when the sum of MAVs from both channels exceeds a preset threshold value 4 healthy subjects participated in these experiments, each of the 6 reference patterns being computed by averagixig 50 measurements (in total 600 segments of the EMG signal lasting 160 ms each were used). With an Euclidean minimum distance classification scheme and short-time training, a 95.3% success rate for 3 DOF was obtained. The computations were all done in real time by the above described hardware.