STUDENT PROJECTS
Patrick Carey - End-to-End Framework for Autonomous Vehicle Simulations
Recently, new research in simulated testing has reduced validation costs of developing autonomous driving technology up to 99.9%. While there is much literature on autonomous vehicles, and utilizing Google Cloud with TensorFlow, there is little research that compiles all the steps and possible solutions that need to be considered for developing models in this framework. The goal of this paper is to utilize Google Cloud’s platform by exploring models in TensorFlow and demonstrating a proof of concept for new advanced driver assist and autonomous vehicle systems. Special emphasis is placed on creating pixel-level labels and prediction of driving images and behaviors using Deep Learning.
In lung cancer diagnosis, it has been more and more common to incorporate Computer Assisted Diagnosis (CAD) softwares to help radiologists reduce workload and make informed decisions. The central piece in this process involves the detection of lung nodules, which encompasses nodule classification, nodule locality detection and pixel-level nodule segmentation. The recent advancement in deep neural networks provides new perspectives in approaching these tasks. In this project, using CT images from The Lung Image Database Consortium image collection (LIDC-IDRI), I applied U-net, a type of convolutional neural network with encoder layer concatenated to corresponding decoder layers, to perform semi-automatic lung nodule segmentation. The result of my experiments showed promising signs of this method, and various suggestions for further improvement are derived from this project.
Tim Zhang - Lung Nodule Segmentation with Convolutional Neural Networks
Shobit Kishore & Ian Wang - 3D object reconstruction from images
The goal of this project is to implement Structure from Motion (SFM) techniques to reconstruct an object in the form of a 3D model from multiple images. The same principles with small tweaks would apply if instead of multiple images from the same camera, different cameras clicked the same object. SFM has been widely used in many applications, such as robot navigation, autonomous driving, and augmented reality.
The goal of this project is to implement Structure from Motion (SFM) techniques to reconstruct an object in the form of a 3D model from multiple images. The same principles with small tweaks would apply if instead of multiple images from the same camera, different cameras clicked the same object. SFM has been widely used in many applications, such as robot navigation, autonomous driving, and augmented reality.
Ray Li & Carlton Smith - Image Retrieval System
For this project, we built a content-based medical image retrieval system for the Lung Image Database Consortium image collection (LIDC-IDRI) using the bag-of-feature approach. The dataset used to extract visual features consists of cropped-out malignant lung nodule patches and CT scan images with benign nodules, and the Scale Invariance Feature Transformation (SIFT) algorithm was applied to extract the visual feature keypoints. After the feature keypoints are extracted, K-means clustering algorithm was applied to group them into more representative feature “bags”. For each search image, by calculating the distance between each of its feature keypoint and the feature “bag”, the image was transformed into a vector representation whose length equals to the number of feature bags and where each element in the vector represents the count of each feature bag it contains. When a new image is used as the query image, its SIFT feature keypoints are extracted and based on the feature bags generated using the database images, the query image is transformed using vector representation as well. By comparing the similarity of the query image vector and database image vectors, a ranked list of images are retrieved from the database.
Stephanie Besser - Motion Analysis of Animal Movement
For the past decade, there has been some research in the area of motion analysis for the betterment of both human and animal movement. Most recently, in 2015, the University of Missouri opened a Motion Analysis Laboratory in the Veterinary Health Center, whereby they assist animals with musculoskeletal and neurologic conditions, as well as support in evaluating treatment and post-surgery progress. The motivation for this project is two-fold: I have always had a love of animals, and I am interested in advancing my knowledge of motion analysis, specifically 3D object detection and tracking of future movements.
​
​