Contributing to AI Development for "Behavior Prediction and Estimation": "Daily Behavior and Natural Actions Video Dataset" Capturing Changes in Daily Actions Now Available

PIXTA Inc. (Shibuya-ku, Tokyo; Representative Director and President: Daisuke Komata; TSE Standard: 3416), which operates the "PIXTA" marketplace for stock photos, illustrations, videos, and music, will begin selling the "Daily Behavior and Natural Actions Video Dataset," comprising 50 video clips and 25 types of actions, as part of its "Image and Video Data Provision Service for Machine Learning." This dataset is not merely a collection of isolated actions but focuses on context-rich behaviors in real-world environments. It includes videos that capture changes in behavior such as starting to move, stopping, changing direction, and hesitation, making it a valuable dataset for AI development related to behavior understanding and prediction.

◆ PIXTA Image and Video Data Provision Service for Machine Learning: https://pixta.jp/machinelearning-dataset

## Overview of the Daily Behavior and Natural Actions Video Dataset

Name: Daily Behavior and Natural Actions Video Dataset

Number of Data Points: 50 video clips
*Customized shooting under specific environments and conditions is also possible based on this package.

Price: 198,000 JPY (tax included)
*Various annotations are available for an additional fee; please inquire.

Shooting Locations: Indoors and outdoors in Japan

Composition: Horizontal to slightly downward viewing angle relative to the subject

Casts: Japanese males and females across a wide range of apparent ages, from infants to seniors

Duration: Approximately 5 to 50 seconds per video clip

Content: This dataset focuses on context-rich behaviors rather than simple actions. It consists of 25 types of actions, including "atypical behaviors."
- Behavioral changes (hesitation, turning back, interrupting actions, etc.)
- Gaze/attention (shifting gaze in multiple directions, fixed gaze, etc.)
- Hand movements (placing objects, changing grip, interrupting actions, etc.)
- States (waiting, lingering, etc.)

Data Format: Live-action MP4 (model release obtained)

Features:
- Available for commercial use
- Permission for machine learning data utilization obtained from videographers

How to Purchase: Please contact us via "Inquiries" below.
https://pixta.jp/machinelearning-dataset

## Intended Uses

This dataset can be utilized in a wide range of fields by corporations, research institutions, and local governments engaged in the development and research of AI models for human behavior prediction and estimation. Specifically, by using "video data," it becomes possible to analyze "changes in behavior over time," which was difficult with still images. The following uses are envisioned:

Development of Behavior Prediction and Change Detection Systems
It can be used as AI model training data for moments when direction or movement changes, such as "turning back" or "interrupting an action (stopping)." By using video data, it can be applied to the development and verification of algorithms that identify and predict subsequent changes in movement.

Analysis of Movement Patterns in Facilities and Outdoors
It can be utilized for analyzing behavior when transitioning from walking to "hesitation (e.g., looking around)" or "stopping" in stores, offices, on roads, or in parks. By using video data, it can be applied to research on analysis models that consider fluctuations in movement occurring along a travel route.

Research on Behavior Recognition Models in Daily Scenes
It can be used as training data for AI to correctly recognize behaviors such as stopping, changing direction, and switching objects in everyday environments like rooms, convenience stores, and offices. By using live-action videos, it can be applied to building behavior estimation models under diverse backgrounds and lighting conditions.

As Fundamental Data for Behavior Recognition and Intent Estimation AI
It can be used as learning data for research and AI development related to "human behavior understanding" in the field of computer vision, including object detection, tracking, and pose estimation with skeletal detection.

## Examples of Industries Expected to Utilize

- Autonomous Driving / Robotics: Development of autonomous vehicles, delivery robots, security robots
- Retail / Stores: Analysis of people's movements in stores, detection of suspicious movements, unmanned stores
- Nursing Care / Medical: Monitoring systems, prediction of falls, AI for dementia care
- Crime Prevention / Security: Advancement of surveillance cameras, automatic detection of suspicious behavior
- Urban Planning / Transportation Infrastructure: Analysis of pedestrian flow in stations, airports, and parks
- Universities / Research Institutions: Research on AI that interprets human behavior

## Background of Provision

PIXTA has previously provided image datasets for machine learning. Meanwhile, we have received requests from various sites for "Japanese video data that captures not just isolated human actions, but a series of changes in daily movements."

Behind this is the acceleration of efforts to socially implement camera image analysis technology in fields directly related to "safety and security," such as crime prevention, monitoring, and autonomous driving. In December 2025, the government approved the "Artificial Intelligence Basic Plan," covering nursing care and crime prevention.