Automatic stitching is quite useful when it comes to stitching full view panoramas. By selecting a number of panoramic shots, all these images can be stitched together into one large beautiful picture! This was achieved using Harris corners as interest points. Once a large number were determined for the images to be stitched, adaptive non-maximal suppression was used to minimize the number of interest points in each image. A bias/gain normalized feature descriptor is extracted for each interest point, and then the descriptors of the images to be matched are compared using the results of 1-NN and 2-NN matching. These matches are further refined with RANSAC and the best homography is selected to stitch the images together.
The original images were taken near the University of Pennsylvania campus, 11 floors up.
Stitched Image :: View
Presentation :: [ PPT ]
A
laser-based navigation and obstacle avoidance algorithm for mobile robots
has been developed. This method allows a mobile robot to navigate an
environment with a large
number of dynamic obstacles and static obstacles (convex as well as
concave). Construction of an efficient testbed has been discussed as well.
Successful navigation is achieved with a global path planner (A*) and a local path planner (potential field navigation function). The laser scanner data is interpreted to distinguish between static and dynamic obstacles and the global path planner computes a path around static obstacles leaving navigation of dynamic obstacles to the local path planner. High quality videos can be downloaded below:
Demo :: View
Report :: [ PDF ]
Simulated World:
world.avi :: (2,186 KB)
Local Minima (navigating with the local planner and without the global Planner):
locmin1.avi :: (2,987 KB)
locmin2.avi :: (2,613 KB)
Examples:
eg1.avi :: (10,454 KB)
eg2.avi :: (21,980 KB)
Using the Gazebo simulator, MATLAB, and a C++ implementation of the A* algorithm, this robot navigates its way through known obstacles after planning a path using feedback linearization.
Simulation :: View
Using MATLAB, and by manually specifying 44 control points, image morphing was achieved with Delaunay triangulation.
Image Morphing Video :: View
Presentation :: [ PPT ]
Using a number of search algorithms, Q-Learning and MDP , we navigate Pacman through mazes in the most optimal manner.
Learning algorithms were implemented in order to identify people based on their facial appearance and a paragraph of news text about the person. For text classification a Naive Bayes classifier with a document model approach was utilized and for face recognition, we used an SVM classifier with a feature-based approach using 45 SVMs.
Project Report :: [ PDF ]
We developed mobile autonomous robots to converge through a maze. These robots were completely designed by us from scratch and were controlled using the TI MSP430 microcontroller unit.
Demo :: View
Report :: [ PDF ]
This mechatronics project involved the design of a pair of devices which autonomously orient toward one another and estimate their separation distance. We used an infrared transreceiver pair to achieve this. The devices were controlled using the MC9S1232 and the TI MSP430 controller units.
Report :: [ PDF ]
Using the Microsoft EBox and MATLAB, we developed a mobile robot capable of autonomous webcam-based control to navigate a slalom using the colors of the poles as a navigational aid. We processed the image using MATLAB and using the segmented image, we located the centroid of the object using BLOB analysis. With this information, the robot was able to navigate through the slalom.