Composed pose retrieval (CPR) enables users to search for human poses by specifying a reference pose and a transition description, but progress in this field is hindered by the scarcity and inconsistency of annotated pose transitions. Existing CPR datasets rely on costly human annotations or heuristic-based rule generation, both of which limit scalability and diversity. In this work, we introduce AutoComPose, the first framework that leverages multimodal large language models (MLLMs) to automatically generate rich and structured pose transition descriptions. Our method enhances annotation quality by structuring transitions into fine-grained body part movements and introducing mirrored/swapped variations, while a cyclic consistency constraint ensures logical coherence between forward and reverse transitions. To advance CPR research, we construct and release two dedicated benchmarks, AIST-CPR and PoseFixCPR, supplementing prior datasets with enhanced attributes. Extensive experiments demonstrate that training retrieval models with AutoComPose yields superior performance over human-annotated and heuristic-based methods, significantly reducing annotation costs while improving retrieval quality. Our work pioneers the automatic annotation of pose transitions, establishing a scalable foundation for future CPR research.

Falls are a major cause of injury among elderly adults. In addition, recent research indicates that elderly adults often fall because of the same individual biomechanical causes. Therefore, real-time individual fall risk assessment is crucial for designing a more effective and personalized fall reduction training program. However, existing methods for fall risk assessment either raise privacy concerns or require multiple wearable sensors. In this paper, we propose EgoFall, a real-time privacy-preserving fall risk assessment system using a commercial tracking camera. EgoFall utilizes a chest-mounted tracking camera and a carefully designed data pre-processing pipeline to acquire the ego-body motion data of the subject. The ego-body motion data is then fed to a lightweight CNN-Transformer model for fall risk assessment. We demonstrate that EgoFall not only outperforms the baseline methods but also has very low computational complexity, which is highly suitable for real-time processing.