Projects

Continuous Human Pose Estimation Using Long Short-Term Memory and Particle Filter

Abstract:

Estimating human pose in a continuous time series has many practical applications. For example, imagine that some time in the future robot would like to interact with human beings, for that robot to meaningfully interact with a human it needs to interpret and anticipate human movements and gestures. Acquiring continuous human pose estimates can also inform specific applications like brain-machine interface; specifically, we can use accounts of human pose data across time to study the relationship between neural signals and human pose. In this thesis, we will focus our work on the continuous human pose estimation in the clinical environment.

There are many existing methods for estimating human pose from camera image, and many of them employ deep learning and convolutional neural network (CNN) architecture, which are widely used in computer vision. However, after estimating possible human poses from a single image frame, might we be able to use the statistical regularity of human movement to improve pose estimation? In this work we demonstrate that by modeling this regularity across time pose estimation can be improved. We demonstrate this by applying a post-processing method to confidence maps of pose generated using existing computer vision methods applied to each frame. Our post-processing method models movement using a long short-term memory (LSTM) network and a particle filter based framework for estimation.

PatientPose: A semi-automated framework for pose estimation in clinical environments

Abstract:

In this paper we describe a semi-automated approach for improving automatic upper-body pose estimation in noisy clinical environments, whereby we build and adapt around an existing joint-tracking framework to improve its robustness to environmental uncertainties. Our approach uses subject-specific convolutional neural network (ConvNet) models learned on a subset of each dataset chosen to maximize feature variance during training time. By adapting to scene lighting changes and by smoothing predicted joint locations through a Kalman filter with fitted noise parameters, the expanded framework yields higher labeling consistency and accuracy compared to similar methods for several hospital patients recorded in complex clinical settings.

Ensemble Classification Algorithms on Fashion Mnist Data

Abstract:

The purpose of the project is to find the best way to classify the Fashion Minst data, which is a Minst-like dataset has ten classes with 60000 training data and 10000 testing data. To achieve our goal, we tried many ensemble classification algorithms as well as studied how can the discriminant learning method improve the classification result. Besides that, we also implemented the Convolutional Neural Network and compared whether the ensemble traditional machine learning method can outperform the deep learning method.

Classification of Iceberg from Satellite Radar Images Using Convolutional Neural Network and Multitask Learning

Abstract:

In this paper, we will use Convolutional Neural Network and multitask learning method to build a binary classifier that can distinguish icebergs from ships. The data we use comes from radar images that are taken from the Satellites over 600 kilometers above the earth. Alone with the images, we also have the angle information when the images were taken. In this project, we utilize the angle information with multitask learning and train a Convolutional Neural Network model to classify whether the image contains an iceberg or ship.