Human-to-Human Interaction Detection

TL;DR

This paper introduces a new task called human-to-human interaction detection (HID), along with a benchmark dataset AVA-I and a Transformer-based model SaMFormer, enabling joint detection, recognition, and grouping of interactive human behaviors in videos.

Contribution

The paper presents the first comprehensive approach to HID, including a new dataset AVA-I with detailed annotations and a novel end-to-end Transformer-based model SaMFormer for simultaneous detection and grouping.

Findings

SaMFormer outperforms existing methods on AVA-I

AVA-I contains over 85,000 frames with interactive group annotations

End-to-end training improves interaction detection accuracy

Abstract

A comprehensive understanding of interested human-to-human interactions in video streams, such as queuing, handshaking, fighting and chasing, is of immense importance to the surveillance of public security in regions like campuses, squares and parks. Different from conventional human interaction recognition, which uses choreographed videos as inputs, neglects concurrent interactive groups, and performs detection and recognition in separate stages, we introduce a new task named human-to-human interaction detection (HID). HID devotes to detecting subjects, recognizing person-wise actions, and grouping people according to their interactive relations, in one model. First, based on the popular AVA dataset created for action detection, we establish a new HID benchmark, termed AVA-Interaction (AVA-I), by adding annotations on interactive relations in a frame-by-frame manner. AVA-I consists of 85,254 frames and 86,338 interactive groups, and each image includes up to 4 concurrent interactive groups. Second, we present a novel baseline approach SaMFormer for HID, containing a visual feature extractor, a split stage which leverages a Transformer-based model to decode action instances and interactive groups, and a merging stage which reconstructs the relationship between instances and groups. All SaMFormer components are jointly trained in an end-to-end manner. Extensive experiments on AVA-I validate the superiority of SaMFormer over representative methods. The dataset and code will be made public to encourage more follow-up studies.