Egocentric Video Search via Physical Interactions

Abstract

Retrieving past egocentric videos about personal daily life is important to support and augment human memory. Most previous retrieval approaches have ignored the crucial feature of human-physical world interactions, which is greatly related to our memory and experience of daily activities. In this paper, we propose a gesture-based egocentric video retrieval framework, which retrieves past visual experience using body gestures as non-verbal queries. We use a probabilistic framework based on a canonical correlation analysis that models physical interactions through a latent space and uses them for egocentric video retrieval and re-ranking search results. By incorporating physical interactions into the retrieval models, we address the problems resulting from the variability of human motions. We evaluate our proposed method on motion and egocentric video datasets about daily activities in household settings and demonstrate that our egocentric video retrieval framework robustly improves retrieval performance when retrieving past videos from personal and even other persons' video archives.

Generating an Event Timeline About Daily Activities From a Semantic Concept Stream

“…We evaluated our proposed method using datasets of ADLs, which are manually annotated with structured event descrip- tions of the ADLs that subjects performed in a house. This dataset has been used in egocentric video retrieval with gesture motions (Miyanishi et al 2016). The dataset consists of motion signals (e.g., acceleration and gyro) and first-person vision videos captured by a head-mounted wearable camera that enables us to capture various ADLs in diverse places.…”

“…Generating Semantic Concepts We made semantic concepts from the signals of wearable sensors. For feature extraction from motion signals and videos, we followed the past work (Miyanishi et al 2016). For the motion-feature extraction, we used acceleration and gyro signals and applied a short-time Fourier transform, where the window width was 75 samples (3 sec when using 25-Hz data) by shifting one sample.…”