Walk in Place Based VR Locomotion System

Kokane, Sanket Suresh

4. term

Vision, Graphics and Interactive Systems, Master

2023

2023-06-02

VR-lokomotion giver ofte cybersyge, hvilket forkorter sessioner og svækker indlevelsen, fordi mange systemer flytter synsfeltet uden at matche kroppens sanser. Denne afhandling undersøger en gå-på-stedet (Walk-in-Place, WiP) løsning, der lader brugere simulere gang, mens de står stille. For at undgå dyre løbebånd eller bærbare sensorer og begrænsningerne ved rent regelbaserede, synsbaserede metoder udvikles et webcam-baseret WiP-system, der opdager gang med Mediapipe Pose-estimering og en dyb læringsmodel, integreret i en Unity VR-prototype. Designet omfatter både manuel bearbejdning af kropslandemærker og LSTM-baseret bevægelsesklassifikation samt en realtidsforbindelse mellem poseinferens og en lokomotionskontroller i scenen. En brugerstudie-opsætning på flade og skrånende terræner sammenligner WiP-prototypen med traditionel controller-baseret bevægelse og vurderer brugeraccept, oplevelse og cybersyge ved hjælp af standardiserede spørgeskemaer. Det medfølgende uddrag dokumenterer motivation, relateret arbejde, metode og testopsætning; konkrete kvantitative resultater fremgår ikke af de viste sider.

VR locomotion often causes cybersickness, shortening sessions and breaking immersion because many systems move the viewpoint without matching bodily cues. This thesis investigates a walk-in-place (WiP) solution that lets users simulate walking while standing still. To avoid costly treadmills or wearables and the limits of purely rule-based, vision-only methods, it builds a webcam-based WiP system that detects walking with Mediapipe Pose estimation and a deep learning model, integrated into a Unity VR prototype. The design explores manual landmark processing and LSTM-based motion classification, and connects real-time pose inference to an in-scene locomotion controller. A user-study setup on flat and sloped terrains compares the WiP prototype with conventional controller-based movement, evaluating user acceptance, experience, and cybersickness using standardized questionnaires. The provided excerpt documents the motivation, related work, methodology, and test setup; specific quantitative results are not included here.

[This summary has been generated with the help of AI directly from the project (PDF)]

Download
View record in AAU Student Projects

Kokane, Sanket Suresh:

• Supervised Multi-Domain GAN for Low Light Image Synthesis (2022)