This project describes the “Scalp Skin Tensiometer” that I have designed and built under the guidance of my mentor, Dr. Poul Nielsen, Professor at Auckland Bioengineering Institute, The University of Auckland, New Zealand.

The instrument measures and displays the tension in scalp skin around the wound area when it has been excised. It gives quantitative information of the scalp skin tension while incision is being closed and sutured, along with the displacement b/w the two edges of the incision.

Firstly, I tell you why it was required to make such an instrument. The human skin basically shows varied mechanical properties…


Ultrasound Image generation
Ultrasound imaging provides a non-invasive and cost effective way to obtain internal images for medical applications and diagnosis of human health. Image generation from ultrasound waves consists of three steps- production of sound waves by piezoelectric transducer, receiving echoes, and formation of image. Strong, short electrical pulses from ultrasound machine make the transducer ring at a desired frequency, which can be anywhere between 2-18 MHz. Then, focussing of sound is done by either the shape of the transducer, lens in front of transducer or a complex set of control pulses from ultrasound machine (also called beamforming), which produces…

Level Sets


The development of active contour models (snakes) results from the work of Kass et al. and they offer a solution to a variety of tasks in image analysis and machine vision. The active contour model, or snake, is defined as an energy-minimizing spline — the snake’s energy depends on its shape and location within the image. Local minima of this energy then correspond to desired image properties.

Snakes do not solve the entire problem of finding contours in images; rather, they depend on other mechanisms such as interaction with user, interaction with some higher level image understanding process, or information…

Pedestrian Tracking


While detecting objects in an image has been getting a lot of attention from the scientific community, a lesser-known and yet an area with widespread applications is tracking objects in a video or real-time video stream. That’s something that requires us to merge our knowledge of detecting objects in images with analyzing temporal information and using it to predict movement trajectories.

This project is inspired from recent YOLOv4 and DeepSORT papers. The project consists of 4 parts: Detection, Estimation, Association and Tracks Handling.

Kalman Filtering

Kalman filtering is an algorithm that provides estimates of some unknown variables given the measurements…


This project was inspired from the paper by Criminisi et al. from Microsoft Research. In the paper, they proposed an approach in which they apply random regression forests to predict the location of anatomical structures in CT scans. They use a data set of 100 torso CT scans, which differ highly in several aspects (e.g. scanner type, resolution, organ size and pose, etc.). For these scans, the 3D bounding boxes for the organs of interest are computed at first to generate the ground truth. …


Q-learning is simple yet powerful off-policy model-free reinforcement algorithm that creates a table of state-action values (Q-values) for the agent that helps the agent figure out exactly which action to perform.

Q-learning Update

Q-learning updates are done as shown above whereas actions are chosen epsilon-greedily i.e. with a policy where with epsilon probability random action is chosen for the next state, and with the rest of probability greedy action with highest Q-value is picked. In order to tackle situations where both state and action spaces are too large in case of Atari games, this Q-learning algorithm needs a boost. We…


Semantic Image Segmentation is a form of dense segmentation task in Computer Vision where the model outputs dense feature map for the input RGB image with same dimensions (height and width) as the input image. Output feature map consists of as many channels as there are classes for each pixel to predict from. Training the model is done by keeping activation for the output layer as SoftMax and loss is computed as categorical cross entropy compared to ground truth segmentation mask. The final prediction mask is then passed through argmax function to bring pixel-wise labels into existence. Hence, the output…

Lohit Kapoor

Machine Vision Researcher

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store