Aman Goyal

Aman Goyal

Graduate Student at Carnegie Mellon University

Selected Projects

Intrusion Detection in Low-Light scenarios

  • A video surveillance based solution for low light scenarios.
  • The motivation behind this solution was to enhance the security of the borders using AI given the rise in terrorism.

  • Low-light images were processed using stacking and averaging techniques.
  • Detection of intruders and their firearms were done using YOLOv4 based model.
  • HRNet was utilized for intruder pose estimation
  • Viola-Jones algorithm was used for intruder face recognition.

The implementation can be found here.

Vehicle Parking Occupancy Detection

  • A parking occupancy status detection solution which utilizes classical computer vision concepts such as Canny Edge Detection and deep learning based object detection model YoloV3’,

The following procedure was followed to derive the solution:

  • Suitable images were found using Open IP CCTV camera websites.
  • Parking zones were detected using Canny Edge Detection and Hough Line Transform.
  • Detection of vehicles took place using YoloV3 model.
  • Vehicles were represented as point objects on the 2D map using homography estimation.
  • Finally the status of the parking zone was detected using Point Polygon Test.

The detailed procedure can be found in this blog.

Deep Learning based COVID-19 classifier

  • COVID-19 is one of the most destructive pandemic humankind has ever faced.
  • It has not only shaken the economy across the world but also made it an extremely difficult task for effective medical help to reach across remote regions.
  • An extremely accurate and lightweight AI model could be a solution for this as it could aid the doctors and medical team especially in remote areas to detect COVID-19.

Hence keeping this in mind, this project aims to create a deep learning based COVID-19 CT-Scan images classifier.

  • In the project, experiments have been done both with as well as without transfer learning.
  • With just 50 training images, MobileNet based model was able to provide an accuracy of 99%.

The implementation can be found here.

Image Denoising with Autoencoders

This project aims to perform denoising of images from Olivetti Faces Dataset by using Convolutional Autoencoders.

  • The images were provided random noise using basic python code.
  • The best results were obtained when the architecture was trained using 1000 epochs.
  • The above image is an example of evaluation done on testing set.

The implementation can be found here.

QuickDraw - Image Recognition

This project was inspired by Google’s QuickDraw and aims to recognize images across 15 categories

  • It is a CNN based image classifier which classifies images across 15 categories.

The implementation can be found here.