Introduction
The purpose of this research project is to develop novel and
innovative, home-based stroke rehabilitation assistive devices. So far,
in this work, we have built an embedded "smart can" (shown on the
right).
In general today, when post-stroke patients complete their formal
therapies, they are provided only with written home exercise programs
prescribed by their therapists, and these are commonly discontinued
after a period of time. In general, compliance with written home
exercise is low, but compliance with home exercise tools has been shown
to be high. To this end, an integrated team of scientists at UB are
working on designing and deploying assistive technology devices, aimed
at improving compliance primarily via objective feedback and
personalization.
We therefore present the computer vision and graphics techniques employed in this research work, such as one-shot object detection, incremental visual tracking, activity recognition, and 3D virtual augmentation. These vision techniques are exploited in the context of enhancing objective feedback on exercise performance and tailoring exercise programs to appropriately challenge participants. For experimental analysis, we record 30 videos of five non-stroke participants performing three pre-specified tasks twice - normally and with jerky movements. We successfully demonstrate the efficacies of the computer vision components of the system in the lab setting and going forward, additional usability tests will be performed to optimize the system to the specific needs of its targeted users.
Awards and Papers
- (Award) Innovative
Micro-Programs Accelerating Collaboration in Themes (IMPACT)
- Internal UB award
Title: Portable Measurement Devices to Provide Feedback and Enhance Self-Management in Chronic Stroke Rehabilitation
PI: Lora Cavuoto; Co-Investigators: Ifeoma Nwogu, Jeanne Langan, Heamchand Subrayan
- Vision-Augmented
Home-based Stroke Rehabilitation
Device
Smriti Jha, Ifeoma Nwogu, Lora Cavuoto, Heamchand Subrayan, Jeanne Langan
(In submission) Elsivier Journal on Computer Vision and Image Understanding (CVIU)
Videos and Images
- Here are the image results from the CVIU submission. First
the performance of the one-shot object detection algorithm:
The results of auto-detection on the first frames of 30 test videos are shown. -
Next are sample videos from the experiments showing two of the exercise tasks;
The first two videos show a participant drinking with normal movements and the virtual rendition of the can motion based on the video.
Next twos video show a participant pouring with jerky movements and its virtual rendition. Click on images to play videos.
-
Videos comparing the tracking results from the incremental visual tracker we used to those from the traditional particle filter-based tracker:
-
Tracking Results: Using IVT [.mov 6.9mb] and using PFT [.mov 6.9mb]
-
-
Information about the IMU sensors embedded in this first generation of the device can be found here.
The accelerometer dataseheet can be accessed here.
Trouble viewing the videos? Please email me if
there are any problems viewing the videos. The
windows media
player should be able to show them on windows. On a Mac, they can be
readily viewed with Quicktime.
Code and Data
To be released....