Realtime Obstacle Detection for VisuallyImpaired PeopleAdvisor(s) Friedrich Fraundorfer David SchinaglInstitute for Computer Graphics and Vision (ICG)Graz University of Technology Austria

Motivation

The Walkassist is a smart shoe developed by an aus-trian start-up called Tec-Innovation to enhance visu-ally impaired peoples quality of life The built-in warn-ing system uses ultrasonic distance sensors to warnusers of obstacles like steps curbs or other peopleIf an obstacle is detected the user is alerted usingacoustic or vibration signals In this project we tryto support the ultrasonic based system by exploitingadditional visual information A camera based systemplaced on top of the toe-cap in combination with deeplearning methods should be used to detect obstaclesand segment the walkable area respectively

Goals and Tasks

The goal of this masterrsquos thesis is to develop a mobilesystem which is capable to detect the walkable areain realtimeThe main parts of this project are

Porting of existing algorithms to a mobileplatform like a virtual reality backpackExtend and improve parts of the pipeline(real-time capability)Measuring the distances to the obstaclesVisualization of the walkable area and thedetected obstacles

This master thesis is offered together with Tec-Innovation and will be paid

Deliverables

Project files (zip)Documentation (pdf)Presentation (pdf)

Prerequisites

PythonC++ programmingInterest in machine learningGPU programming (optional)

Advisor Contact

Friedrich Fraundorfer (fraundorfericgtugrazat)David Schinagl (davidschinaglicgtugrazat)

Master Thesis Computer Vision Aerial Vision