As a part of master thesis project, ES-EKF SLAM with quaternion kinematics was implemented to reconstruct the stereocamera trajectory and thereby reconstructing the object. Use of ES-EKF SLAM allows to have linear quaternion kinematics.

As a part of master thesis project, a stereocamera system was designed which is able to reconstuct objects with sub-millimetre depth resolution. Concept of positive and negative pattern projection by means of optical projector is evaluated provide robustness to specular reflection. As a consequence of wide baseline, DAISY desriptors is calculated.

In this project, MPNet—a motion planning network, the technique employed by Qureshi et al. is used. This network receives environment information as point-clouds, as well as the robot’s initial and desired goal configurations and recursively calls itself to bidirectionally generate connectable paths. MPNet is implemented using Tensorflow, an end-to-end open-source machine learning platform with Python.

3D reconstruction of the face using measurements captured by 3DMD system. For this purpose, Computer Vision Toolbox is used. The image shows SURF keypoint correspondences.

Concepts used while controlling the robot (e.g., twist, geometric jacobian etc.) have deep geometric meaning and are coordinate independent. This borrows concept of forward kinematics followed by differential kinematics, position control. Using these concepts position and velocity control laws are implemented.

Detecting brain hemorrhage using deep learning approach and investingating the use of saliency maps to understand how sensitive is a network to a small change of grey levels in a given region.