Education
Masters Computer Science Aug 2016
University of North Carolina at Chapel Hill
Coursework:
Computer Vision, Virtual Worlds, Robotics, Machine Learning, Algorithms, Files and Databases, Compilers
University of North Carolina at Chapel Hill
Coursework:
Computer Vision, Virtual Worlds, Robotics, Machine Learning, Algorithms, Files and Databases, Compilers
Bachelor of Science (Honors) Electrical Engineering June 2013
School of Science and Engineering
Lahore University of Management Sciences(LUMS), Lahore
Coursework:
Digital Image Processing, Computer Networks, Data Structures,Data Acquisition and Interfacing, Computer Organization,
State and Parameter Estimation
School of Science and Engineering
Lahore University of Management Sciences(LUMS), Lahore
Coursework:
Digital Image Processing, Computer Networks, Data Structures,Data Acquisition and Interfacing, Computer Organization,
State and Parameter Estimation
Work Experience
I worked as a research assistant with Professor Henry Fuchs in the Telepresence group at UNC Chapel Hill. We work on developing Telepresence systems with Head Mounted Displays (HMDs). Current HMDs have a large form factor and so occlude most of the face. Along with my colleagues, I worked on removing the HMD from the reconstruction and replacing it with the user's face. In addition, by putting small cameras inside the HMD and accounting for the lens distortion, I was able to recreate the user's eye movements and blinking from inside the HMD.
I worked as a Computer Vision Intern at Organic Motion in the summer of 2015. I worked under the supervision of my manager Dan Amerson to develop a wide area prop tracking system. The image processing required for the algorithm was performed on the GPU to ensure realtime performance comparable to their OpenStage product. This allowed me to learn and use GPGPU programming and DirectX while sharpening my knowledge of Computer Vision techniques. I presented the system to the team at various stages of development and incorporated their feedback. I had a wonderful experience working with my colleagues there who provided useful feedback and aided my learning process.
I was a Research Assistant at the Computer Vision Lab at LUMS in Lahore from Nov 2013 to June 2014. Here, I worked on the Human Pose Estimation problem in images. I collaborated with my predecessor, Ali Hassan to extend the Pose Estimation Framework that he had developed. Our method used SVM classifiers to compute likelihood for the position and orientation of each limb. We used this with a prior and a PCA based pose validation scheme to compute the pose Estimates for the whole body. In particular I collaborated with Ali to come up with search space reduction techniques to reduce likelihood computations. By leveraging Kinematic constraints, we limited the search region to 35% of the image. Finding local maxima using Mean Shift leads to a further reduction by 50%. This project allowed me to put my machine learning knowledge into practice which I had acquired through online courses. I am thankful to Dr. Murtaza Taj for his support and guidance through this project. This work was published in International Conference on Image Processing
After finishing my Bachelors degree I was a Research Intern at Robotics Research Lab at TU Kaiserslautern for two months. I worked on improving indoor localization using visual-inertial sensor fusion. The goal was to combine a slower but more accurate Visual tracker with a faster and less accurate inertial tracker to I used the RGBDSLAM algorithm with Kinect to get the pose estimate. This was slow but more accurate. I developed an Extended Kalman Filter to fuse measurements from an Inertial Measurement Unit (IMU) onboard a smartphone with the pose estimate obtained from the RGBD-SLAM algorithm. to Through this project, I gained experience of various theoretical state estimation techniques and practical experience of the Robot Operating System and Eigen C++ library. I was supervised by Dr. Abubakr Muhammad and Dr. Karsten Berns during this internship. I am grateful to Dr. Abubakr for teaching me the fundamentals of Probabilistic State Estimation techniques.
During the spring of 2012, I worked as an Undergraduate Research Assistant at CYPHYNETS Lab with Dr. Abubakr Muhammad. I was involved in conducting outdoor experiments with Kinect with RGBD-SLAM for in low sunlight conditions. In addition I was involved in developing robotic platforms for testing navigation algorithms and telepresence applications. This internship helped me to develop a holistic understanding of robotic systems. This resulted from working at all levels; from Robot Operating System for higher level control, microcontrollers to generate the PWM signals and finally to motor driver circuits.
Skills
Languages: C++, C, Python, Java, MATLAB, MIPS Assembly
Libraries: Tensorflow, Keras, ROS, OpenGL, OpenCV, PCL, Microsoft DirectCompute, Oculus Rift SDK, Eigen
Sensors: Laser Scanners, RGBD sensors, IMUs
General Skills: Deep Learning, Natural Language Processing, Computer Vision, Image Processing, GPGPU programming, Machine Learning, Robot Programming, Interfacing
Platforms: Windows, Linux
Database: MySQL
Libraries: Tensorflow, Keras, ROS, OpenGL, OpenCV, PCL, Microsoft DirectCompute, Oculus Rift SDK, Eigen
Sensors: Laser Scanners, RGBD sensors, IMUs
General Skills: Deep Learning, Natural Language Processing, Computer Vision, Image Processing, GPGPU programming, Machine Learning, Robot Programming, Interfacing
Platforms: Windows, Linux
Database: MySQL
Publications
- A. Khalid, A. Hassan and M. Taj, 'Efficient 2d Human Pose Estimation Using Mean- Shift', in IEEE International Conference on Image Processing 2014 (ICIP 2014), Paris, France, 2014.
- Taj, Murtaza, Ali Hassan, and Abdul Rafay Khalid. "2D Human Pose Estimation and Tracking in Non-Overlapping Cameras." Human Behavior Understanding in Networked Sensing.Springer, 2015. Print.