Comparison of pre-trained models for object recognition used in performing ADLs
Yulith V. Altamirano-Flores, Iván Gónzalez, Irvin Hussein López-Nava, Luis Cabañero, Jesús Fontecha · IEEE Access · 2026
The identification of Activities of Daily Living (ADLs) is a fundamental task in various applications, including smart surveillance, health monitoring, and assistive technologies for individuals with special needs. In particular, identifying ADLs carried out in the kitchen environment—such as those related to food preparation and consumption—is essential, as the performance of these activities provides valuable insights into users’ functional autonomy and well-being. Given that such activities predominantly occur in kitchen settings, this environment represents a strategic context for the implementation of intelligent recognition systems. This study explores the use of pre-trained deep learning models, specifically YOLOv8-based models, to enhance object recognition performance in real-world kitchen scenarios. This work represents a foundational step toward addressing the specific challenges of kitchen-related ADL identification in future steps, focusing on the recognition of objects used or interacted with in the aforementioned ADLs, as a preliminary procedure that will provide useful information for subsequent ADL identification. The models have been trained and optimized using the three-stage methodology presented in this work. In particular, their performance has been evaluated using the Mean Average Precision metric. The results demonstrate a substantial improvement in the recognition of key objects in kitchen environments, providing a significant step forward for further integration of object contextual information into ADL identification systems. Future work will focus on including more object categories, retraining the models using the proposed methodology, and adapting them for robust object recognition in multi-frame video sequences.