Optical flow applications for measuring microvascular erythrocyte velocity
- Yang Cheng
- Oct 27, 2022
- 1 min read
Updated: Oct 27, 2022
Can we build a wearable medical device to measure the risk of stroke smartly?
One goal of this project is to combine the data collected from the eyes of the patients and computer vision techniques to develop new medical hardware to assist doctors in measuring the risk of stroke. Some people may want to use a graph neural network to tackle the case on the deep learning side. However, things can be easier with traditional computer vision techniques.
The first problem is how to detect the blood cells in the image. Shi-Tomasi corner points define the moving objects. Then Lucas-Kanade optical flow is used to classify their pixels as the same object and measure the speed. The result shows in the picture above.
The second problem is finding the clusters of blood cells in the video. Since clustering may indicate that the patient's blood cells are sticky, leading to a high risk of stroke. I designed an algorithm called temporal k-means to find cluster cells. Instead of using information from one frame, I added time as the third dimension of the measurement. So we can see objects which move together in a video. This method has another version in which we use speed as the third measurement.
The initial goal has been changed in the process of the whole project. The published paper is part of the middle product.
Paper link:
Comments