Position: Research Scholar
Location: Harvard Biorobotics Laboratory, Harvard John A. Paulson School of Engineering and Applied Sciences
Period: 2014 – 2015
In this work, customized high-speed algorithms are developed for the segmentation and feature extraction from the images of two arbitrary positioned
cameras. The image processing algorithms in this research involve three main steps of (a) preprocessing; (b) tip point detection; and (c) centerline points extraction. The preprocessing procedure comprises of removing lens distortions from the images, cropping, thresholding and removing the blue back- grounds, removing the noise, and ﬁnding the position of the origin points by ﬁnding the location of the middle point of the the most top-right and most top- left white points in the images. A customized algorithm was designed to ﬁnd the position of the tip point as the conventional methods like Hough transform for ﬁnding circles was found to be sensitive to noise resulting in the detection of features with different properties for the images from identical views. To reproduce the centerline of the catheter from the images, several techniques can be employed including SOM, detecting ridges using a distance transform, calculating the Voronoi diagram, thinning layer by layer erosion, and etc. While in practice these techniques work well, to lower the execution time, a customized algorithm was developed. This algorithm begins with applying a mask containing a circular arc centered at the origin point to the binary images of the isolated catheter. The proposed algorithm allows the processing rate of around 200 fps or even more.