SyncWISE: Window Induced Shift Estimation for Synchronization ofVideo and Accelerometry from ...
summary
Presents a concrete algorithm called SyncWISE that aligns wearable camera video and wrist accelerometer streams by estimating window-induced time shifts, implemented and evaluated on 163 videos totaling 45.2 hours from 21 participants. Reports 91% accuracy for synchronization within 700 milliseconds tolerance, outperforms prior methods even with high data loss, and reproduces state-of-the-art results on the CMU-MMAC dataset, with implementation details and validation numbers included.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:22Time Synchronization Between Video and Accelerometry
- 0:46Sense2Stop: In-the-wild Smoking Cessation Study
- 1:13Data Screening and Annotation
- 1:34Key Challenges
- 2:11Data Preprocessing - Chest-worn Accelerometer
- 2:28Data Preprocessing - Video
- 2:53Method Overview
- 3:08Approach
- 3:58Experiments - Synthetic Sense2Stop-Sync
- 4:26Experiments - Original Sense2Stop-Sync
- 4:53Experiments - CMU-MMAC
- 5:14Discussion
related talks
17mMadEye: Boosting Live Video Analytics Accuracy with Adaptive Camera Configurations
34mMoving in Sync: Self-supervised learning of n-human interactions
18mFitness motion recognition with Android wear
27mFrom minutes to seconds: Capillary auto-alignment with python and opencv
06mWrite, Attend and Spell: Streaming End-to-end Free-style Handwriting Recognition Using Smartwatches
20mPathSync: Multi-User Gestural Interaction with Touchless Rhythmic Path Mimicry
21mSelf-Calibrating Head-Mounted Eye Trackers Using Egocentric Visual Saliency
06mDetecting Stressful Social Interactions Using Wearable Physiological and Inertial Sensors