On a cold day in early December, the lobby of Goodwin Hall was crowded with posters, students, and interested members of the public, all talking animatedly about computer vision. Teaching machines to see was the subject of a fall 2015 ECE course, taught by Assistant Professor Devi Parikh. More than 80 undergraduate and graduate students completed the course, which culminated in a final project that asked them to apply their lessons to real world problems.

Aravind Premkumar (MSCPE '17) and Sonal Pinto (MEng CPE '17)

Flappy Bird is a simple game where the user navigates a bird through a series of obstacles. Tap-to-Jump games like Flappy Bird and TRex Runner the game you can play in Google Chrome when there is no Internetrely on hand-eye coordination. Aravind Preshant Premkumar and Sonal Pinto outsourced the playing to a computer. Working off webcam-recorded videos of each game, they used MATLAB to build algorithms for image acquisition and processing, and a capacitive actuator, which worked like a button to register taps.

Divya Bala (MSCPE '16), Joshua Stuckner (Ph.D. MSE '19), Swazoo Claybon (MSEE '17)

In order to diagnose and treat diseases like cancer, doctors and researchers must first identify cells, often by analyzing microscope images. Manually, it can become critically time consuming. Swazoo Claybon, Divya Bala, and Joshua Stuckner used a micrograph and a magnified image of cells in an attempt to provide information about cell type and number. In their first model, they trained a high-level image classifier, and developed class specific detection algorithms. Their second approach involved a generic cell detector followed by classification.

Matt Bender (Ph.D. ME '17), Mincan Cao (MEng CPE '16), Yu Wei (MEng EE '17), and Shaoxu Xing (MSME/MSEE '16)

Bat flight mechanics make excellent inspiration for flapping wing Micro-Air Vehicles. Matt Bender, Mincan Cao, Yu Wei, and Shaoxu Xing developed a three-step method for identifying marker locations and tracking points.First, they isolated the bat's body from the image background by using a frame differencing approach, which produces a bounding box containing the bat.Next, they computed filter bank responses within the bounding box to determine the location of features.Finally, they implemented a Kalman filter-trackingalgorithm to track the points.

Brian Cesar-Tondreau (Ph.D. ME '20), Alfred Mayalu (Ph.D. CPE '18), Joshua Moser (MSME '16)

This team looked into a wearable device that would "look" around a lab and automatically identify tools. As a proof-of-concept for this idea, they developed software that would be able to recognize and classify a set of hand tools using feature descriptors such as scale-invariant feature transform (SIFT), and a machine-learning system like a Support Vector Machine (SVM).They found that the best results came from using SIFT in conjunction with a border angle feature space. However, even with a model that had an accuracy above 70 percent, some tools, like hammers, were still classified incorrectly.