No abstract

No abstract

No abstract

This study presents a new neural network model for recognizing manual works using body-worn accelerometers in industrial settings, named Lightweight Ordered-work Segmentation Network (LOS-Net). In industrial domains, a human worker typically repetitively performs a set of predefined processes, with each process consisting of a sequence of activities in a predefined order. State-of-the-art activity recognition models, such as encoder-decoder models, have numerous trainable parameters, making their training difficult in industrial domains because of the consequent substantial cost for preparing a large amount of labeled data. In contrast, the LOS-Net is designed to be trained on a limited amount of training data. Specifically, the decoder in the LOS-Net has few trainable parameters and is designed to capture only the necessary information for precise recognition of ordered works. These are (i) the boundary information between consecutive activities, because a transition in the performed activities is generally associated with the trend change of the sensor data collected during the manual works and (ii) long-term context regarding the ordered works, e.g., information about the previous and next activity, which is useful for recognizing the current activity. This information is obtained by introducing a module that can collect it at distant time steps using few trainable parameters. Moreover, the LOS-Net can refine the activity estimation by the decoder by incorporating prior knowledge regarding the order of activities. We demonstrate the effectiveness of the LOS-Net using sensor data collected from workers in actual factories and a logistics center, and show that it can achieve state-of-the-art performance.

The development of a machine-learning-based human activity recognition (HAR) system using body-worn sensors is mainly composed of three phases: data collection, model training, and evaluation. During data collection, the HAR developer collects labeled data from participants wearing inertial sensors. In the model training phase, the developer trains the HAR model on the collected training data. In the evaluation phase, the developer evaluates the trained HAR model on the collected test data. When the HAR model cannot achieve the target recognition accuracy, the developer iterates the above procedures by taking certain measures, including collecting additional training data, until the re-trained model achieves the target accuracy. However, collecting labeled data for HAR requires additional time and incurs high monetary costs. In addition, it is difficult to determine the amount and type of data to collect for achieving the target accuracy while reducing costs. To address this issue, this paper proposes a new method that predicts the performance improvement of the current HAR model, i.e., it determines the level of performance improvement achievable by re-training the HAR model with additional data, before collecting the additional data. Thus, the method enables the HAR developer to establish a strategy for additional data collection by providing advice such as "If labeled data for the Walking and Running activities from two additional participants is collected, the HAR accuracy of the current HAR model for Walking will improve by 20%." To achieve this, a neural network called AIP-Net is proposed to estimate the improvement in performance by analyzing the feature space of the current HAR model using the proposed entropy-based attention mechanism. The performance of AIP-Net was evaluated on eight HAR datasets using leave-one-dataset-out cross-validation.