A lightweight model for indoor object detection in unstructured scenes based on joint attention and prior knowledge in the context of home rehabilitation

Abstract

Environmental perception is the core support of intelligent assistive systems in home rehabilitation scenarios. The unstructured characteristics of home rehabilitation scenarios pose significant challenges to indoor object detection. In addition, current research on indoor object detection mainly focuses on developing more complex network architectures. To address these issues, this paper proposes an efficient indoor object visual recognition model (IOVRM). IOVRM includes three innovative plug and play modules: multi-scale feature extraction module (MFEM), joint attention mechanism (JAM), and multi-scale feature fusion module (MFFM). These three modules respectively improve the computational efficiency, key feature extraction ability, and multi-scale feature fusion ability of the model, thereby enhancing the model's ability to recognize indoor objects in complex environments while reducing model complexity. In addition, a training strategy based on prior knowledge was proposed. This training strategy significantly improves the model's generalization ability by training the model on an existing dataset to obtain prior knowledge and using it as part of the weights. The experimental results show that IOVRM achieves an effective balance between performance and efficiency, with corresponding numerical results of mAP50, mAP50-90, Params, and FLOPs of 64.0% and 45.8%, 10.64, and 25.7, respectively. IOVRM provides a solid foundation for task execution of home service robots. This study not only achieved the collaborative optimization of high-precision detection and lightweight deployment, but also supplemented the theoretical system of indoor object detection in the context of home rehabilitation, providing practical solutions for rehabilitation assisted intelligent devices and home health monitoring systems.