Department of Electrical Engineering

Permanent URI for this communityhttps://repository.cuilahore.edu.pk/handle/123456789/18

Browse

Filters

2020 - 20212

Settings

Author: search.filters.author.Dr. Khuram ALi, Assistant Profesor [Supervisor]

Search Results

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Deep Learning Based Hand Gesture Recognition Using
    (comsats university islamabad lahore campus, 2021) Rija Sohail,; SP19-REE-017; Dr. Khuram ALi, Assistant Profesor [Supervisor]; LHR TP 7469
    In last couple of decades, much work has been done on Human Activity Recognition (HAR). HAR is a broad research area and excelled in many applications such as security, health care, gaming, intelligent environments and activity of daily living. One of the main applications of HAR is Hand Gesture Recognition (HGR). Hand gesture recognition is a complex classification problem. Previous studies show that different sensor technologies and different classification approaches have been used for gesture recognition. But still certain aspects need to be addressed so that the robustness and reliability of the gesture- based models can be improved. This research work is comprising of four traditional machine learning models SVM, KNN, Random Forest, Decision Tree and three deep learning models RNN-LSTM, CNN-LSTM and CNN. These seven models are developed and implement for the hand gesture data acquired from IMU. These models are then evaluated on the basis of different parameters. The analysis of result shows that deep learning-based models are clearly the best choice to be used for HGR systems.
  • No Thumbnail Available
    Item
    A Hybrid Framework For Attitude Estimation Of
    (Publisher COMSATS University Islambad Lahore Campus, 2020) Talal Ali Khan; SP18-REE-005; Dr. Khuram ALi, Assistant Profesor [Supervisor]; LHR TP 7471
    Inertial Measurement Unit (IMU) sensors are being actively used for motion tracking or attitude estimation due to their low weighted structure and cheap cost. IMU is a fusion of gyroscope, accelerometer, and magnetometer. There are many applications of IMU sensors. Some of which are gesture recognition, estimation of vehicular motion and orientation of objects, etc. Mostly gyroscope and accelerometer are used to estimate the attitude of any object where magnetometer provides the heading and course information. The gyroscope provides short-term stability because of inherited bias. Due to this bias an error starts to occur as the difference between actual measurement and expected measurement of the attitude tends to increase with time. This error can be large as 70 degrees/hour. This problem can be solved by using accelerometer in fusion with gyroscope. Accelerometer that measures accelerations acting on a body due to gravitation and other exterior forces. It is relatively easy to estimate attitude when there are no outliers but in presence of external interference its measurement is contaminated which creates jitters and lag in measured values. As the external forces are always present that’s why it never accurate. This is why we use these sensors in conjunction so; they can complement each other to give accurate results. To solve this problem, we have introduced an effective technique in which complementary filter is used to give accurate attitude estimate of the body in the presence of prolonged dynamic conditions and gyro bias. Two filters are used in cascade form where one filter deals with precise attitude estimation of the motion of a rigid body and external acceleration through synchronicity of accelerometer and gyroscope and the other filter deals with gyro bias estimation, this enables the proposed architecture to be used in a different application with minimum calibration. A novel threshold-based method is used for external acceleration detection so that external accelerations can be dealt efficiently. This will help in changing the confidence level on the measurements of accelerometer in case of prolonged accelerations and also compliments attitude estimation. Second filter is designed in a way that it estimates bias in the by using attitude estimation and accelerations profile and then the bias is removed from the system. The architecture is also compared with different estimators existing in the literature under different dynamical conditions on different datasets.