Measuring Biometric Data from Mobile Phone Video

In collaboration with Markitech and YourDoctorsOnline and with Mitacs funding, BAM lab researchers developed a remote video based biometric data measurement system (incorporated in the Veyetals product) which uses smart phone video to measure heart rate, heart rate variability, oxygen saturation, stress level, blood pressure, and a separate contactless sensor to measure body temperature.

Chat Analysis and Summarization

Chats offer challenges as these are often short responses, may not be a complete sentence, can have a lot of noise (because of machine voice to text translation or typing error), and have domain specific terminology and phrases that are used in general conversations. Our work on chat summarization focuses on medical chats between real patients and real doctors over a medical advising platform to summarize the past chats to assist the doctors with providing an informed medical advice.

Hierarchical Classification of Images for Online Shopping Market

In collaboration with a startup company, BAM lab researcher (undergraduate) developed an efficient hierarchical image classification algorithm to categorize clothing using camera for online shopping market.

Identifying Reportable Diseases for Ministry of Health and Long Term Care

Created a cognitive data analytic and decision support pipeline for classifying noisy medical data based on schema and ministry specified guidelines using both a rule based and a neural network approach to identify reportable critical diseases [Lucas Rychlo].

A voice-controlled e-commerce application using IBM Watson

Reviewed state-of-the-art speech recognition systems, proposed a taxonomy and developed a speech enabled e-commerce website to increase accessibility of such websites using IBM Watson speech-to-text API and cloud services [Kandhari et al., 2018].

Autonomous Cars

Our CUTRIC funded project focuses on extracting key features from vehicular data for efficient V2V and V2I communication.
In an exploratory study, undergraduate students (Daisy Barrette and Alex Weatherhead) constructed two different versions of autonomous cars, one using ultrasonic sensors and deep learning models, and another using camera and deep convolutional networks, where the cars were trained to autonomously drive by sensing track boundaries. The model cars were made using raspberry pi with bluetooth connections, arduino uno and commercially available cheap sensors.

Vehicle Detection using Computer Vision

In this project, we implemented two different deep learning models based on YOLO series algorithms and compared their performances in detecting vehicles such as cars and trucks for self-driving cars to ensure road safety. In detecting the vehicles within 50 meters, the two variations of our models, YOLO and YOLOv3, achieved 93.75% and 96.88% mean recall, and 82.50% and 97.92% F1-measure on a two-way road with a detection speed of 19 fps and 74 fps respectively.

Computer Vision

Video Object Tracking: Leveraged the YOLOv2 architecture with a 2-D recurrent LSTM to implement a predictive video object detector and tracker (POD) [Gasmallah et al, 2018]. Another project focused on detection and measurement of sphroids, a model for cancer cell study.

Wearable Sensors and IoT

Fall Detection using Wearable Sensors: A deep learning model was implemented and trained offline using the public MobiAct dataset and later deployed in a streaming IoT data analytics framework for fall detection using MbientLab sensor MetaMotion R [Ajerla et al., 2018].