Drones serve as the primary camera-mounted platform to collect high-resolution images of buildings from various viewpoints for 3D neural radiance fields (NeRF) reconstruction. However, the flexibility of drones in lowering flight heights to supplement richer building details and thus improve the reconstruction quality of NeRF has not been explored. Given the limited power of a drone and the rapidly changing nature of outdoor scenes, it is necessary to quickly predict an optimal supplementary lower flight height after the drone finishes capturing at a default height. The main challenges involve the implicit relationship between supplementary flight heights and reconstruction quality, the complexity of prediction task with little prior information, and the tension between the need for real-time decision making and the resource constraints inherent to on-drone task execution. In this work, we develop an end-to-end system pipeline with an offline-online decoupling feature. We first design a model architecture that embeds supplementary flight heights to predict the reconstruction quality of NeRF using images captured at the default height. To enhance model generalization ability in a cost-effective manner, we offline train an ensemble of multiple models over the samples constructed from sandbox buildings using building-level cross validation. During the online serving phase on the drone, the most appropriate model is selected based on feature matching between real and sandbox buildings to determine the optimal supplementary flight height. We build a testbed and evaluate on 16 sandbox buildings and 8 real-world buildings. Evaluation results demonstrate that our pipeline accurately identifies the optimal supplementary lower height in a few minutes and improves reconstruction quality.
Machine learning has become a prominent technique used in wireless sensor networks (WSNs) in the past 10-15 years. Because of their high accuracy, adaptability, and potential to lesson computational overhead compared to existing mathematical algorithms, machine learning techniques are excellent for dynamic environments. Within machine learning, there are many different techniques to consider, such as supervised learning, unsupervised learning, reinforcement learning, and others. This paper will focus on Q-learning, and its advantages in WSN selection algorithms. We propose a Q-learning-based sensor selection framework to optimize node scheduling while providing k-coverage by learning policies that balance coverage and remaining energy. In particular, sleep scheduling algorithms are investigated and the Q-learning developed policy is compared with an existing mathematical algorithm via simulations. To conclude, the advantages and disadvantages of machine learning with resource constrained WSNs is discussed.
The fusion of multi-modal information, such as images and LiDAR scans, is instrumental to maximize the performance of many computer vision tasks in next generation systems and applications. However, supporting fusion necessitates considerable effort, challenging the availability of computing and communication resources in edge systems. This work addresses this challenge by maximizing resource efficiency in systems where mobile devices collect multi-modal sensor data and use dynamic multi-branched DNN models to adapt inference to the operating context. To tune the overall system response to the context (e.g., weather conditions), we propose a dual-scale control approach: centralized orchestration of spectrum resources, and distributed individual device-level control of the execution path of the dynamic DNN fusion models. The control agents are driven by a novel context-aware decision-making method combined with game theory, named Context-Aware Network Slicing Auction (CANSA), which optimizes DNN inference performance, network slicing, and energy consumption. The decision-making performs such optimization by: (i) selecting data and features that best fit the current context; (ii) deciding on the appropriate DNN model complexity, including the use of multi-modal sensor fusion techniques for better data integration, and (iii) deploying these models on the most appropriate nodes (local nodes or edge servers). Results, obtained using real-world multi-modal data, show that CANSA surpasses conventional allocation methods by up to 52.3% in terms of inference task success rate.
Door Access Control (DAC) plays a pivotal role in balancing security and convenience in modern infrastructures. However, current vision-based DAC systems exhibit limitations including privacy concerns (e.g., facial data leakage), performance degradation under suboptimal lighting conditions, high computational overhead and system costs. While RFID and Bluetooth-based alternatives exist, they exhibit vulnerabilities to attacks including replay attacks, signal cloning, and eavesdropping. Recent advances in visible light sensing and backscatter communication have enabled promising opportunities for secure, low-power access control systems with sub-dollar hardware costs. In this paper, we propose ViKey, the first visible light backscatter-based DAC system that utilizes polarized birefringence to generate 3D position-dependent color patterns as keys, enabling robust and contactless authentication. We design and implement a ViKey prototype using commercial off-the-shelf (COTS) components, with a tag cost of less than $0.2. Real-world experiments show that our current ViKey prototype can achieve an average authentication accuracy of 90.5% at 0.5m with our best patterns. These results demonstrate the effectiveness of our low-cost visible light backscatter technology for future smart DAC applications.
In WiFi and many other sensing applications, researchers have adopted the approach of representing each activity class with a small portion of data to address user variability across different environments, thereby enabling model generalization. This solution raises yet another challenge - limited training samples for each activity class, particularly for a considerable number of classes to be classified. In this paper, we first introduce a visionary architectural layering architecture that enables the end-to-end lifecycle of multi-sensor federated learning, paving the way for realizing federated AI-based sensing. We identify three challenges, e.g., data heterogeneity, scalability under non-IID conditions, and few-shot learning. According to the architecture, we showcase a lightweight, general workflow that utilizes a federated learning framework to address the few-shot learning challenge across various classifications in WiFi Channel State Information (CSI). We analyze this workflow and identify two key factors that most significantly affect its performance (i.e., the aggregation method and the model architecture). Based on the analysis, we propose three network models for CSI-based activity recognition: CSI-AlexNet, CSI-ActNet, and CSI-ResNet. Through extensive performance evaluations, our experimental results show that the proposed approach achieves an accuracy of 97.97% on CSI-ActNet while maintaining low computational demands, making it well-suited for real-time fine-tuning on edge devices with constrained resources.
