In mobile crowdsourcing (MCS), the task information released by task publishers and the sensed data submitted by workers may expose their privacy, while the rapid growth of MCS imposes increasing data processing pressure on cloud platforms and mobile devices. To address these challenges, a bilateral privacy protection scheme based on adaptive location generalization and grouping aggregation is presented in this article. The scheme uses federated learning as a framework and utilizes edge computing to reduce the data processing burden on cloud platforms and mobile devices. This article proposes the adaptive location generalization algorithm (KM-ALG) and a real task location release mechanism based on the RSA algorithm to protect the task location privacy of the task publisher. For workers’ privacy protection, the lightweight multiple perturbation algorithm based on localized differential privacy (LDP-MP) proposed in this article is used to protect workers’ data privacy. Aiming at the problem of data quality loss caused by perturbation, a perturbation elimination mechanism based on homomorphic encryption technology is proposed. In order to prevent workers’ sensed data from leaking location information, a grouping aggregation mechanism is used to destroy the correspondence between workers and submitted data, thereby protecting workers’ location privacy. In addition, a task allocation scheme adapted to task location privacy protection is also proposed. Finally, the effectiveness of the proposed algorithm is verified through experiments on multiple real data sets.