Recently, several AR systems have been developed and used in various fields. However, in most AR systems, there are some restrictions caused by the usage of AR marker or location information. In this research, in order to solve these problems, AR information system that can recognize object itself based on deep learning was developed. In particular, this system was constructed using client-server model so that the machine learning can be updated while operating the system. In addition, the method of gamification was introduced to gather the learning data automatically from the users when they use the system. The prototype was applied to the AR zoo information system and the effectiveness of the proposed system was validated in the evaluation experiment.