Several outbreaks of COVID-19 caused by imported cases have occurred in China following the successful control of the outbreak in early 2020. In order to avoid recurrences of such local outbreaks, it is important to devise an efficient control and prevention strategy. In this paper, we developed a stochastic discrete model of the COVID-19 epidemic in Guangzhou in 2021 to compare the effectiveness of centralized quarantine and compulsory home quarantine measures. The model was calibrated by using the daily reported cases and newly centralized quarantined cases. The estimated results showed that the home quarantine measure increased the accuracy of contact tracing. The estimated basic reproduction number was lower than that in 2020, even with a much more transmissible variant, demonstrating the effectiveness of the vaccines and normalized control interventions. Sensitivity analysis indicated that a sufficiently implemented contact tracing and centralized quarantine strategy in the initial stage would contain the epidemic faster with less infections even with a weakly implemented compulsory home quarantine measure. However, if the accuracy of the contact tracing was insufficient, then early implementation of the compulsory home quarantine with strict contact tracing, screening and testing interventions on the key individuals would shorten the epidemic duration and reduce the total number of infected cases. Particularly, 94 infections would have been avoided if the home quarantine measure had been implemented 3 days earlier and an extra 190 infections would have arisen if the home quarantine measure was implemented 3 days later. The study suggested that more attention should be paid to the precise control strategy during the initial stage of the epidemic, otherwise the key group-based control measure should be implemented strictly.