KidRisk: Benchmark Dataset for Children Dangerous Action Recognition

2026-06-23

AI Summary

This study addresses the challenge of recognizing hazardous actions by unsupervised children through the introduction of KidRisk, the first large-scale multimodal dataset comprising 2,500 videos and 10,000 images. The authors propose a vision–language joint modeling framework that integrates deep learning with contextual semantic understanding to significantly enhance the discrimination of risky child behaviors in complex real-world scenarios. Experimental results demonstrate that the proposed method achieves an accuracy of 83.53% on general child action classification and a markedly higher accuracy of 96.14% specifically for dangerous action recognition, substantially outperforming conventional deep learning approaches.

Abstract

Children are naturally energetic, and during their spontaneous activities, they often encounter potentially dangerous situations, especially when lacking parental supervision. Identifying actions that pose risks plays a crucial role in ensuring their safety. This paper build a novel challenging dataset, namely KidRisk, including 2,500 short videos of children's actions and 10,000 images for dangerous action of children. We also introduce a benchmark on our newly constructs dataset and find that traditional deep learning models demonstrated limited effectiveness on these tasks. Therefore, we develop vision-language based baselines with exceptional context understanding of visual information. Our proposed methods achieved an accuracy of 83.53% in classifying children's actions and 96.14% in recognizing children's dangerous actions, significantly outperforming traditional approaches. These results confirm that vision-language models are not only feasible but also highly effective in detecting hazardous actions, contributing positively to safeguarding children's safety.

Problem

• children dangerous action recognition
• child safety
• action recognition
• risk detection
• video understanding

Innovation

• vision-language model
• dangerous action recognition
• children safety
• benchmark dataset
• context understanding