By leveraging a network of Raspberry Pi devices equipped with standard cameras, DeepDish addresses the need for real-time, accurate monitoring of people's movements within a space. This approach not only aids planners and designers but also enhances public safety and data-driven decision-making for communal spaces.

System Overview and Impact

The DeepDish project implements a novel system of edge devices, primarily utilizing the Raspberry Pi equipped with standard camera peripherals. This network, poised for integration with LoraWAN, aims to revolutionize real-time data monitoring in urban and building contexts. The core objective is to furnish stakeholders, including urban designers, fire marshals, and the general public, with critical data on occupancy and movement patterns. This data not only informs safety measures and design improvements but also fosters a data-driven approach to communal space management. The integration into a Digital Twin of a local site marks a significant leap towards a comprehensive, virtual representation of physical spaces, enhancing the predictive and analytical capabilities of urban planning and management.

Experimental Setup and Results

Our experimental setup comprises a Raspberry Pi network capable of processing image sequences to track multiple objects simultaneously. The deployment focuses on a pilot project aimed at monitoring entry and exit points of a building, using this data as a precursor to a larger-scale analysis of public space utilization. Preliminary results demonstrate the system's efficacy in providing real-time event and statistics publishing over a secure, low-bandwidth network. The flexibility of the setup allows for adjustments in camera placement and field-of-view settings to optimize person-counting accuracy, even with the system's limitation to processing 2-3 frames per second.

Privacy Considerations

DeepDish is designed with a strong emphasis on privacy. The system's architecture ensures that image data is processed locally on each Raspberry Pi, with only anonymized, aggregated data transmitted over the network. This approach minimizes the risk of personal data breaches, adhering to privacy laws and regulations. Furthermore, the project's commitment to privacy extends to the deployment phase, where cameras are positioned to collect necessary data without intrusive surveillance, ensuring a balance between valuable data collection and respect for individual privacy.

Accuracy and Performance

The performance evaluation of DeepDish revolves around its ability to accurately track multiple objects through machine learning algorithms. Despite the resource constraints of the Raspberry Pi, the system demonstrates a high level of accuracy in object detection and tracking. By employing models such as MobileNet and DeepSORT, DeepDish achieves efficient real-time tracking with minimal latency, suitable for live monitoring applications. The system's performance underlines the potential of off-the-shelf hardware in executing complex computational tasks, providing a cost-effective solution for multi-object tracking with promising scalability and adaptability to various environments.

Conclusions and Future Directions

DeepDish highlights the potential for scalable, cost-effective multi-object tracking solutions in urban and building contexts. Future work will focus on expanding the system's capabilities, including broader deployment and integration with additional data sources for enriched analytics.