In order to ensure the safety of the elderly in indoor environment, and to detect dangerous situations and provide rescue support in time, a set of indoor risky behavior monitoring and early warning system for the elderly was developed, which can identify common risky behaviors of the elderly and give early warning. For the abnormal prone and falling, two common risky behaviors of the elderly, the RGB camera was used to collect human posture information and extract key feature points of human skeleton.. Support vector machine was used to classify all kinds of risky behaviors and further combine the behavior status and location information to provide remote warning and emergency call to the relevant personnel. The results show that the true classification rate of risky behavior recognition by the system is 97.14%. The system can accurately monitor and warn indoor risky behaviors of the elderly, thus effectively slowing down the expansion of accidental injuries and ensuring the home safety of the elderly.