SCDC

Abstract—Sparse Mobile CrowdSensing is an efficient data collection paradigm that recruits participants to gather data from partial spatiotemporal regions and leverages inherent correlations among these data to infer the remaining uncollected data. However, enabling accurate inference requires participants to upload sensitive spatiotemporal information, which poses significant privacy leakage risks. Traditional methods address these risks by obfuscating the uploaded spatial data, but this often compromises inference accuracy. To tackle this challenge, we propose a novel spatiotemporal obfuscation strategy that obfuscates collected data to more critical spatiotemporal regions and adjusts the values based on spatiotemporal correlations, ensuring high inference accuracy under privacy constraints. Specifically, we employ an active learning-based exponential mechanism for spatial obfuscation. After spatial obfuscation, an association model is applied to capture spatiotemporal relationships and adjust data values. Then we use a Fourier transform-based switch operation for temporal obfuscation, both designed to guide data towards regions most beneficial for inference, to ensure high-quality completion while maintaining privacy constraints. Experimental evaluations demonstrate that our strategy significantly enhances data completion quality in urban sensing tasks, such as environmental monitoring and traffic management, while providing robust privacy protections.