Investigating Deep fusion under Convolution Neural Network for Autonomous cars, 2016
Object in aerial uav images for pipeline maintenance (Proposal submitted), 2016
The major aim is to prevent human vandalisation of the Nigerian oil and gas pipeline. The system will serve as a means of preventive maintenance.
2D/3D fusion for face modeling
This is a 3D digitalization project on the state of art review on different fusion techniques for face modelling and their evaluations. We have grouped fusion approaches into score based, feature based and asymmetric based fusion techniques. At the end, we have choose the fusion techniques that best suites 3D face modeling in real-world security scenarios such as video surveillance system
Least Square Conformal Maps ( LSCM ) for Automatic Texture Atlas Generation
Computer Graphics: In this project, I and my colleque implemented the Least Squares Conformal Maps ( LSCM ) for Automatic Texture Atlas Generation based on bruno levy's paper. Our project comprises of three main steps which include (1) mesh segmentation, (2) Lscm Parametrization, and (3) Packing
Each above steps is a huge research fields on its own. We developed a new algorithm for our segmentation approach and also implemented D-chart algorithm for segmentation. I have included the report and presentation below.
Text Box
3D reconstruction, Structure from Motion ( 3D Scanner )
This is a full project on implementing 3D shadow scanner using OpenCV. The idea of this project is to understand mathematical details and terms in the field of computer vision. Litearture and Matlab code on this approach was presented on here. We followed the same approach of the author except that we implemented it under OpenCV. This project helps understand mathematics of triangulation, Camera and projector Calibration and 3D reconstruction from intersection of planes.
Note : Feel free to contact me for our written code in OpenCV, I also think this was the first time the approach was implemented in C/C++ and openCV as at that time.
Text Box
Face Recognition using Principle Component Analysis (PCA) [pdf]
There is an attached file below for detailed explanation, This project was based on Principal Component Analysis for Face Recognition, This was homework like project to understand better how concepts of Eigenvectors, Eigenvalues, Singular Value Decomposition (SVD) and Principle Component Analysis (PCA) can be used to solve real world problems. The dataset used for the project was provided by the instructor. It contains the faces at different views of myself and my classmates. We had to develop a face recognition system using PCA in MATLAB based on Matthew Turk and Alex Pentland paper "Eigenfaces for Recognition" ( link ).
Wavelet based image compression
Exploration of 4D MRI Blood-Flow using Stylistic Visualization
This is a literature review.
Other projects
French demography survey Report
Object Tracking using Mean Shift Algorithm
November, to December, 2011
Classification using a Kohonen Network
September to November, 2011
Crack Detection in Industrial Tube (Late 2011)
I worked on this throughout my third semester of masters, My summer internship and My Project under Polarization/ Infrared module was related to this and I try and come up with different ideas as I can during my third semester to improve the result I got. I worked on this throughout as a personal project mainly on weekends as a contribution and test to my proposed algorithm for Industrial tube inspection.
Object Recognition, Biometrics Application (Early 2011)
Face and Object Recognition (Biometrics Application), Probability and Graph method approach for existing holistic approaches, I was also fascinated alot by Local Binary Pattern (LBP), I worked a lot on LBP for Face, Gesture, Gender, recognition of objects.
COMPUTER NETWORK
Implementation of ISO Standards on Risk Managements, Office Machines Information technology (June,2010 to December, 2010)
Nagios Exploration for Computer Networks Monitoring(January, 2010 to June, 2010)
Segmentation for Computer assisted Laparoscopy
Segmentation is one of the main step to any computer vision task. In this thesis, We are interested in 3D reconstruction of Deformable surfaces from Laporoscopic video ( Monocular video ), Segmentation is neccessary for better analysis of the Laparoscopic Video, and later help in the 3D reconstruction but in this application, segmentation approach must be accurate and fast to be used in real-time. Result of our segmentation will be integrated into already developed 3D reconstruction algorithm of laparoscopic video using shading cues. finally, The segmentation results will be used to learn the Bidirectional reflectance distribution function (BRDF) of the surgical tool and organs which will serve as a reliable information for 3D reconstruction of this laparoscopic video.
Maximum likelihood approach for object detection in noisy image
Application for detection of cracks in industrial tube.
This is a project in collaboration with an industrial company under oil and gas tube "confidential"
SPIHT (Set partitioning in hierarchical trees) for Image Compression
This is under Image Compression Module, We implemented the SPIHT algorithm to encode and decode image at the other end. We compared its efficiency to other methods. This algorithm contains two step which are the wavelet part and the Spiht coding part. We study further the use of different wavelets and its effects on image for compression using this algorithm. its based on the paper of SPIHT (Set partitioning in hierarchical trees) method by [Said and Pearlman 96]. The report and presentation of this project can be found in the attached area of this page.
Text Box
3D Mesh Handling
This project is based on creating and extending data structures to handle 3D triangular mesh objects and also the results rendering using OpenGL . Main task of the projects includes Finding Incoherent Points in the mesh, boundary edges of the mesh, nth neighborhood points and faces of any vertex in the mesh, Voronoi Region, mean curvature and normal of each vertex, Geodesic distance between any two given vertex in the mesh. Here is the paper used for proper guidance in the implementation.
Text Box
Affine Structure from Motion [pdf]
Reconstruction of three dimension object from multiple views is a huge research field in computer vision. There have been different methods proposed for reconstruction using different images. We can interpret features in 3D world based on perspective projection. Ideas on perspective projection tells us that parallel lines in 3D world appear to converge to a vanishing point which gives information about 3D. For affine images, they have orthographic projection to preserve parallelism such that when parallel lines in the 3D world are projected on to the image plane, they remain parallel. In this project, we implemented the reconstruction of 3D affine structure from multiple affine images using Factorization method by C. Tomasi and Kanade. This reconstruction was done without the need to know the intrinsic parameters of the camera. We made use of database documents provided by the instructor to test the results .
Text Box
Real Time Image Processing using FPGA ( Lossless Image Compression ) [ppt]
This is a small project that introduce us to how image processing algorithm can be done in real time . We only implemented Image compression algorithm which was calic prediction using FPGA. We made use of ISE design suite, VHDL language. I have included the report only. Another sub-project in this module was based only on Research literature reviews of video compression H.264 implementations in FPGA.
Evaluation of Blood Flow in the Thoracic aorta from cine-MRI
Software development for Medical application to analyse 3D blood flow