Chhavi Sharma

• Image Processing Bot

  Objective: To build an autonomous robot that performs object detection and tracking solely using monocular vision. Description: This project focuses on how effective vision alone can be used as a tool for object finding, navigation and collision avoidance. We aimed at building an autonomous robot with a camera input that performs object detection and pathfinding in order to pick up trash soda cans around the university campus. More details about the visual component of the project can be found in our project paper ‘Object detection and pathfinding using monocular vision’ that later got published in a conference journal sponsored by IEEE.Link

• Traffic Jam Detection

  Objective: To build a simple traffic jam detection application using camera input. Description: Build a traffic jam detection program from a video-feed for the course “Digital Image Processing”. It employs fundamental image processing techniques such as erosion, dilation and thresholding for object detection, and, background subtraction and mean shift tracking for detecting the absence of movement. It was written in Matlab and used it’s image processing libraries.

• Human Detection and Tracking

  Objective: To automate the collection of an image database of walking humans from a video feed. Description: In this project, I built a program to automatically collect a database of images of walking humans for further uses such as GAIT biometric analysis. The videos were recorded on the university campus. HOG (Histogram of Oriented Gradients) features were used to train an SVM (Support Vector Machine) classifier for upright walking human detection and Mean-Shift algorithm was used for their tracking. Thi program was written in Python and used open source libraries such as OpenCV.

• Complete

  Objective: To build a text prediction application implementing Text Correction, Completion and Suggestion at every input interrupt. Description: This text predictor uses basic concepts of Natural Language Processing. Suggestion probability was predicted using an N-Gram prediction based model up to trigrams. Probability estimation was done using Markov Chain Rule. A static vocabulary was used which could be changed into self-learning library later. Since the vocabulary library was static, correction or completion was applied to unknown words. We used a FIFO cache for adaptive prediction.

• Wireless Charger

  Objective: To build a prototype of a tabletop wireless charger, resembling a miniature version of the MIT ‘Witricity’ device. Description: This prototype uses inductive charging based on the principle of electromagnetic induction. It requires two circuits: a transmitter and a receiver. An alternating current is passed through the transmitter coil, generating a magnetic field. This, in turn, induces a voltage in the receiver coil which can be used to power a mobile device or charge a battery. The AC voltage circuit of choice was a slightly modified Royer oscillator.