Computer Vision Enthusiast
Computer Vision Enthusiast
Projects
Object Tracking
November 2019 - December 2019
Implemented four different motion detection algorithms, viz. Simple Background Subtraction, Simple Frame Differencing, Adaptive Background Subtraction and Persistent Frame Differencing. The results of these algorithms on various short videos were compared and analyzed to obtain the best performing algorithm in different natural conditions.
MOTION DETECTION • BACKGROUND SUBTRACTION • FRAME DIFFERENCING
Image Blending
September 2019 - October 2019
Blended two images by down-sampling two 512 x 512 images into 8 x 8 images by Gaussian blurring the images at each step and also creating a Gaussian and Laplacian pyramid simultaneously. Blended the images using a mask and up-sampled it successively by using the Laplacian Pyramid to retain the features till we obtain the original image dimensions.
IMAGE BLENDING • DOWN-SAMPLING • UP-SAMPLING
Image projection and 3D reconstruction by Triangulation
October 2019 - November 2019
Projected the 3D joint locations of the Taiji dataset into 2D image coordinates taken from two cameras of different parameters. Constructed the epipolar lines in the images and mapped them to corresponding epipolar lines in the image in the other camera. The 3D points were reconstructed from the two 2D views using Triangulation and the results were compared to the original 3D points.
CAMERA PROJECTION • EPIPOLAR LINES • TRIANGULATION
Augmented Reality
March 2019 - April 2019
The objective was to implement Augmented Reality. Used the COLMAP software on a set of images to generate a 3D reconstruction of the scene. Implemented the RANSAC algorithm on the generated points to find a dominant plane in the scene. Placed a virtual box on the dominant plane generated. Performed 3D to 2D projection of this scene to view the box in each of the original images from different angles in 2D.
AUGMENTED REALITY • COLMAP • RANSAC
Face Recognition
February 2019 - March 2019
The aim of this project was to achieve a faster and efficient way of Face Recognition. The 'FACE' dataset was augmented using various transformations like rotation, adding noise etc. Augmented Variance Ratio (AVR) was used to select the best subset of features by a Feed Forward approach to reduce the dimensionality. Different classification Techniques such as Fischer's projection, Deep Learning techniques such as developing CNNs with different number of layers and filters were used on this transformed dataset and the results were compared.
DATA AUGMENTATION • BEST FEATURE SELECTION • CONVOLUTIONAL NEURAL NETWORKS
Kalman Filtering
April 2019 - May 2019
Estimated the location of 12 different sets of joints in the human body over time, given the measured joint locations at each instant. The measured estimate is very rugged and the aim of this project was to use a Kalman filter to obtain a smooth estimate of the joint locations over time.
KALMAN FILTER • 12 SETS OF JOINTS • POSITION ESTIMATION
CNN Based Novel Approach for Real Time Object Detection and Grasp Orientation for a Prosthetic Hand
Jan 2018 - May 2018
Aimed towards assisting physically challenged people with their day to day activities. Used CNNs instead of the traditional Electromagnetic signals to recognize and classify objects. A self-trained YOLO algorithm was used for object detection and grasp classification in real time.
CONVOLUTIONAL NEURAL NETWORKS • YOLO ALGORITHM • GRASP ORIENTATIONS
Robotic Path Finder
October 2017 - November 2017
Built a robot that can autonomously navigate through a grid by avoiding perceived obstacles. Used the Lego Mindstorms EV3 kit to build the bot and the Probabilistic Road Map and Image Processing algorithms to detect and avoid obstacles