Crowdsourced navigation is becoming the prevalent automobile navigation solution with the widespread adoption of smartphones over the past decade, which supports a plethora of intelligent transportation system services. However, it is subjected to Sybil attacks that inject carefully designed adversarial GPS trajectories to compromise the data aggregation system and cause false traffic jams. Successful Sybil attacks have been launched against real crowdsourced navigation systems, yet defending such critical threats has seldom been studied. In this work, a novel deep generative model based on Bayesian deep learning is devised for Sybil attack identification. The proposed model exploits time-series features to embed trajectories in a latent distribution space, which serves as a basis for identifying ones generated by Sybil attacks. Case studies on three real-world vehicular trajectory datasets reveal that the proposed model improves the performance of state-of-the-art baselines by at least 76.6%. Additionally, a hyper-parameter test develops guidelines for parameter selection, and a fast training scheme is proposed and assessed to boost the model training efficiency.