Benefiting from the sparse representation of channels in the delay-Doppler (DD)-domain, orthogonal time-frequency space (OTFS)-based visible light communication (VLC) systems can compensate different channel impairments to detect transmitted symbols with the knowledge of channel state information (CSI). However, in prior OTFS-based VLC works, the effective channel estimation method has not been fully explored yet. In this paper, we derive the relation equation between the channel impulse response (CIR) and the received DD-domain pilot signal subblock to provide an efficient channel estimation method based on sparse Bayesian learning (SBL) algorithm. Numerical results have shown that the proposed embedded pilot-aided DD-SBL channel estimation scheme significantly outperforms other conventional methods in terms of estimation accuracy and bit error rate (BER) performance for practical scenarios.