Purpose Different kinds of sensors such as accelerometers and gyroscopes have been used for inferring, predicting, and monitoring human activities for various kinds of applications, including human-computer interaction, surveillance, smart home, health care, and security. In this study, we present a novel and robust method to recognize human activities, including resting, squat, and stepper exercises, solely from photoplethysmography (PPG), which is a non-invasive, simple, and low-cost opto-electronic technique that takes measures from the skin surface. Methods The features were extracted in raw PPG segments by Hilbert transform and then classified by the k-nearest neighbor, naive Bayes, and decision tree algorithms. Results The proposed method was successfully applied to the data set recorded from seven subjects and achieved an average classification accuracy rate of 89.39% on the test data. The smaller standard deviation results proved that the proposed method was robust, and the detection of human activities can be effectively performed by Hilbert transform features and decision tree classifier. Conclusions This PPG-based approach could provide human-activity information in addition to monitoring heart rates and early screenings of various atherosclerotic pathologies, such as cardiovascular and hypertension diseases.