Unlocking the Power of UWB and Deep Learning for Accurate Indoor Positioning
The field of human activity recognition (HAR) has witnessed significant advancements, driven largely by the rapid evolution of Internet of Things (IoT) device technology, particularly in the realm of personal devices. This study delves into the utilization of ultra-wideband (UWB) technology for tracking the daily paths of inhabitants within home environments, leveraging the power of deep learning models.
UWB technology offers a promising approach to indoor positioning, estimating user locations through time-of-flight and time-difference-of-arrival methods. However, the presence of walls and obstacles in real-world environments can significantly impact the precision of these techniques, reducing their overall accuracy. To address these challenges, we propose a fingerprinting-based approach that harnesses the received signal strength indicator (RSSI) data collected from inhabitants while performing their daily activities in two residential settings (60 m$^2$ and 100 m$^2$).
By comparing the performance of convolutional neural networks (CNNs), long short-term memory (LSTMs), and a hybrid CNN+LSTM model, as well as incorporating Bluetooth technology, we aim to uncover the most effective solution for accurate indoor positioning. Additionally, we evaluate the impact of the type and duration of the temporal window (future, past, or a combination of both) on the model’s predictive capabilities.
Our findings demonstrate a mean absolute error close to 50 cm, highlighting the superiority of the hybrid CNN+LSTM model in providing precise location estimates. This level of accuracy facilitates the seamless integration of location data into smart home systems, enhancing the overall experience and enabling more advanced human activity recognition in multi-occupancy settings.
The Evolution of Indoor Positioning: Challenges and Opportunities
The field of human activity recognition has evolved significantly, driven by the rapid advancements in IoT device technology, particularly in personal devices such as smartphones and wearables. These devices are equipped with a wide array of sensors, including accelerometers, gyroscopes, Bluetooth, Wi-Fi, and GPS, which capture various types of information about the user’s environment and activities.
In addition to wearable sensors, ambient sensors, such as cameras, acoustic, and thermal sensors, have also been utilized to collect environmental data in specific locations. By analyzing the data collected from these sensors, researchers and developers have been able to generate valuable insights and solve complex problems in the realm of ubiquitous computing, human-computer interaction, and human behavior analysis.
One of the critical challenges in the domain of HAR is the identification of specific individuals performing activities, particularly in scenarios involving multiple inhabitants, such as in multi-resident apartments. To address this complexity, indoor location tracking methodologies have become essential, enabling the accurate identification and monitoring of trajectories for both individuals and objects within indoor settings.
Navigating the Complexities of Indoor Positioning
Indoor positioning has become a crucial technology in various applications, including smart homes, healthcare, and assisted living. Multiple technologies can be used for indoor tracking and positioning, including radio-based (e.g., UWB, Wi-Fi, Bluetooth), optical (e.g., video camera, infrared), magnetic (e.g., magnetic strength), and acoustic (e.g., ultrasound) methods.
Despite the importance of indoor positioning, several technical gaps persist in the field. One significant issue is the lack of extensive datasets collected in real-world environments, as most experimental setups and evaluations are conducted in controlled laboratory settings, which do not accurately represent the complexities and variabilities of actual indoor spaces.
Historically, indoor positioning systems have primarily relied on Bluetooth low energy (BLE) technology due to its low cost and ease of deployment. BLE is widely integrated into wearables and smartphones, which can act as tags, ambient beacons, or anchors collecting received signal strength indicators (RSSI) values. However, the variability of BLE signals and the dynamic nature of indoor spaces, influenced by furniture and other obstacles, require a high density of beacons to achieve accurate positioning.
Fingerprinting: A Promising Approach for Improved Indoor Positioning
Fingerprinting, a common technique used in location systems, involves creating a map of signal strengths at various locations and using this map to estimate positions. Although effective, this method requires a substantial amount of labelled data, and the performance of BLE can be inconsistent due to signal fluctuations.
In recent years, UWB technology has emerged as a promising alternative for indoor positioning, offering higher precision than BLE, as it allows for coordinate-based localization rather than area-based localization. However, UWB technology also faces significant challenges that affect its performance. Signal obstruction is a primary concern, as UWB signals are susceptible to attenuation and multipath effects caused by physical obstructions such as walls and furniture, leading to reduced accuracy in location estimation.
To address the problem of non-line-of-sight (NLOS) conditions and reduce the impact of the number of anchors required, the fingerprinting method plays a crucial role. By correlating the real-world coordinates with the RSSI values received from each beacon, fingerprinting can provide accurate location estimates even in complex environments, mitigating the limitations of UWB in obstructed settings.
Leveraging Deep Learning for Robust Indoor Positioning
This work is motivated by the need to develop robust, accurate, and cost-effective indoor positioning systems to track entities (persons or objects) within indoor environments. Traditional methods often struggle with signal obstruction, environmental variability, and high infrastructure costs. To address these challenges and improve precision, we focus on radio-based technologies, specifically UWB and BLE.
Our approach involves deploying anchors throughout the facility to receive signals from tags attached to the entities. Initially, the user maps their location within the environment, and these labels are collected and correlated with the signals received by the anchors. These data are sent to a server for processing, where the model learns from the labelled data. Once trained, the system can accurately calculate the location in real-time of the entity using only RSSI signals, reducing costs and facilitating deployment in real environments.
Evaluating UWB and BLE for Indoor Positioning
We have evaluated UWB deployments, including a comparative analysis with Bluetooth for inhabitant location. By comparing UWB with BLE, we aim to determine the most suitable technology for different scenarios. While UWB offers high precision with coordinate-based localization, BLE can be sufficient for applications requiring area-based localization. This comparison helps in selecting the appropriate technology based on the specific needs of the environment.
Fingerprinting techniques have been integrated instead of trilateration or triangulation to deduce the cost of anchor deployment. Learning from the RSSI measurements in an instant of time is proposed in conjunction with others previously saved by the sliding window method. Deep Learning models with convolutional neural network (CNN) and long short-term memory (LSTM) for regression of the user’s location have been evaluated.
A Comprehensive Case Study in Real-World Environments
An ad hoc case study in two flats and different configurations has been deployed, including the ground truth labelled in real-time by the inhabitant using a tablet application. The case studies have been designed to be deployed in a real environment with dynamic tracks representing the daily human activity of each inhabitant under natural conditions.
A comprehensive evaluation and comparison of several machine and deep learning techniques (CNN, LSTM, LSTM+CNN, random forest, and support vector machine) was conducted to identify the most accurate method for predicting the user’s location. Additionally, the influence of window type (past, future, and past+future) and window size was assessed.
Achieving Accurate Indoor Positioning with RSSI Fingerprinting and Deep Learning
Our results demonstrate a mean absolute error close to 50 cm, highlighting the superiority of the hybrid CNN+LSTM model in providing accurate location estimates, thus facilitating its application in daily human activity recognition in residential settings. The performance of the deep learning models, coupled with the RSSI-based fingerprinting approach, outperforms traditional methods, showcasing the potential of this solution in addressing the challenges of indoor positioning.
By leveraging the strengths of both UWB and BLE technologies, along with the adaptability of deep learning models, we have developed a robust and cost-effective indoor positioning system that can accurately track inhabitants in complex home environments. This innovation paves the way for enhanced smart home experiences, improved human activity recognition, and more personalized assistive technologies for the benefit of residents.
Conclusion and Future Directions
This study has demonstrated the effectiveness of using UWB technology and deep learning models for tracking daily paths in home environments. By integrating RSSI fingerprinting and leveraging the power of CNN and LSTM architectures, we have achieved a high level of accuracy in location estimation, with a mean absolute error close to 50 cm.
The insights gained from this research highlight the potential of UWB and deep learning in addressing the challenges of indoor positioning, particularly in multi-occupancy settings. As we continue to refine and expand this technology, we envision its seamless integration into smart home systems, enabling more advanced human activity recognition and personalized assistance for residents.
Moving forward, we aim to explore the deployment of this architecture in real-world, 24/7 operational environments to gather comprehensive data and further enhance the models. Additionally, we plan to investigate edge computing approaches, where UWB scanning is integrated with wearable devices, potentially reducing infrastructure requirements and improving system flexibility.
By continuously pushing the boundaries of indoor positioning technology, we can unlock new possibilities for smart home experiences, healthcare monitoring, and assistive living solutions, ultimately improving the quality of life for individuals within their residential environments.
