Occupants and their activities are important in computer-aided systems such as energy management or energy simulation in residential buildings. Numerous studies have extensively examined detailed household activities at the population level using time-use surveys. Despite the considerable influence of contexts on occupants' activities, a limited number of studies focus on individual households due to the constraints imposed by private policies on collecting occupants' activities. This article proposes an approach to detect detailed occupants' activities (e.g., cooking, entertaining) based on measurements in specific households. Firstly, a camera-free mobile application is proposed to support the data collection process due to the restriction of cameras. Then, relevant features are determined for each activity. Finally, Bayesian networks were built to represent human-understandable relationships between determined features and related activities. An instrumented dwelling with five individuals is tested, and a 4-month dataset is used to evaluate the model quality. Results show that the model can estimate most of the studied activities (e.g., cooking, washing dishes) with a F1-score between 0.77 and 0.92. The approach is better than the statistical approach with time-use surveys in building household activity profiles, which is important in energy simulation and verification. In addition, this approach is replicable for different households, providing them with context-specific and highly useable information to enhance smart energy management systems and occupants’ energy behavior.