Research
My primary research interest is in Computer Vision with emphasis on object and category recognition, tracking, scene and activity analysis. I am also interested in the areas of Pattern Recognition, Machine Learning and Statistical Signal Processing.
Following are some of the projects in my research profile:
3-D Depth Reconstruction from Single Still Images
Working on the development of an interface to reconstruct depth from single still images and temporal cues for a diverse range of scenarios. This project expands to three semesters of the Masters’ program with literature review, feature extraction, fusion and benchmarking to be completed by the end of second semester in France.
Ground Truth Information
Worked in IMAGINE lab at Ecole des Ponts, ParisTech on a tool for automatic, semi-automatic or manual creation of precise image features. Its main goal was to create ground truth data to test automatic feature detectors and descriptors. Features can be of different kinds: points, line segments, regions, etc. All features come with precision information, in particular regarding location in the image.
Human Tracking Across Multiple Non-overlapping Cameras
This work presents approach for an automated surveillance system which performs human detection and tracking across multiple non-overlapping cameras. Emphasis is put at single camera level where motion based segmentation is achieved using optical flow estimation. Feature matching and region-based shape descriptors are used for tracking. The proposed approach then extends feature and region-based shape matching across fields of view (FOV) of multiple non-overlapping cameras in order to obtain inter-camera correspondences. Evaluation of the system was performed in different scenarios which verify that the proposed method can efficiently perform video motion tracking with high processing speeds.
Implementation and Evaluation of Standard Visual Category Recognition Approach Using State-of-the-art Descriptors
The image/video database retrieval problem involves finding in the database, instances of multimedia content that is similar to the content of interest, specified by the user. Visual category recognition also referred to as concept detection is one of the most important tasks in image/video indexing and is the main focus of the project. It aims at annotating videos and images using a vocabulary defined by a set of concepts of interest including scenes types (outdoor, vegetation etc), objects (airplane, car etc), events (people-marching etc) and certain named entities (person, place etc). A standard approach to visual category recognition has been established in the community. This approach involves local descriptor computation, vector quantisation via clustering, structured scene or object representation via localised histograms of vector codes, similarity measure for kernel construction and classifier learning. The aim of this project is to implement and then evaluate this approach using state-of-the-art method descriptors such as OpponentSIFT, RGSIFT on Pascal VOC 2007 and then ways to combine them.
Kernel tracking of density-based appearance models is implemented in this project for real-time object tracking applications. First a ROI, i.e., the region of interest is selected in real-time to create a model. Then the matching and locating of the search object is achieved by using mean-shift algorithm. Experimental results show that this method can find perform object tracking with adaptation to scale and translation, robustness to noise and handling of occlusion.
Computer Vision Education in Developing Countries like Pakistan - A survey of Tools & Projects in Undergraduate Studies
Vision is conceivably the most important of the human senses. It permits perception and understanding of the world. Machine vision is the study of methods and techniques whereby artificial vision systems can be constructed and usefully employed in practical applications. Artificial vision is a complex task because the nature of machine vision is different from the human optical system. Emerged in the late 1950s and early 1960s; artificial vision still continues to evolve. However, during the last ten years, it has evolved from research to an implementable technology. Higher processing speeds, greater memory capacities, improvements in optical systems fueled the development of artificial vision. This survey work studied tools and projects which can used for education of Vision in developing countries with experiments and study performed in Pakistan – a developing country.[View Online]I also worked on electronic automation during my undergraduate program. Here are some of the results related to that:
Design and Implementation of an Autonomous Robot for Line Following and Load Dispatching
Security concerns have become an important issue in the present world. For this purpose, a robot based mechanism is presented to cope with the security issues. A robotic manipulator is implemented based on the mechanism in a catapult; highly flexible steel is used as a tool for throwing defensive elements like shells, etc. It can be remotely controlled via RF transmission as well as pre-programmed to patrol its programmed routes. Microcontroller is used which takes input from passive infrared sensors. The simple design is cost-effective and accurate. The work focused on electronic design as well the mechanical assembly including the interface of both.
Pharmaceutical Medication Device With Supervisory Control and Data Acquisition System
Pharmaceutical Medication Devices are used to fill, count, package & deliver medicines ensuring an automated and efficient batch production. Different types of devices are used but the theme here is to increase the efficiency by introducing Supervisory Control and Data Acquisition System (SCADA) which overcomes human intervention and increases the overall efficiency of the pharmaceutical process. Hardware design of an efficient batch production device is also presented along with the mechanical aspects.
Firmware and Hardware Design of a Novel Door Locking System
A low cost electronic system has been designed which can control different locking mechanisms. Hardware design of an electromechanical lock is also presented in this paper. Microchip® PIC microcontroller is the brain controlling all the operations. Low operating power consumption, small circuit, non-volatile memory operation using on-chip Electrically Erasable Programmable Read Only Memory, flexible mechanical design and user friendly interface are the salient features of this security system.[View Online]