Aim The goal of this project is to contribute effectively to the overall objective of preventing loss of autonomy, by providing home care for older adults, that is efficient, useful and acceptable.Addressing the major public health problem of the fall of older adults, an interdisciplinary projectfounded by the French Research Agency has been set up in order to propose a low-cost system thatcan be implemented in the homes of the elderly. This device, based on depth and/or low costthermal sensors is aimed, on the one hand, to detect falls and, on the other hand, to prevent the riskof falling by analyzing the activity of individuals. MethodsTwo passive acquisition systems for fall detection and activity monitoring have been explored: 1) apair of depth and thermal cameras with processing methods for people tracking using particle filtersand activity recognition using Deep Learning and; 2) a stereo pair of low resolution (80x60 pixels)thermal images with several processes including machine learning-based fall detection and activityrecognition. The sequence of activities are then analyzed in order to establish the routine life of theelderly. Modification of this routine can be a sign of frailty. ResultsLearning datasets have been created in 3 different locations with several people using bothacquisition modalities. On these datasets, the machine learning-based fall detection from a stereopair of thermal images reached an accuracy higher than 0.9 for fall detection. Concerning theactivity recognition, we reached an accuracy higher than 0.95 on depth/thermal acquisitions andthan 0.75 using only one low-resolution thermal image. For the activity analysis phase, since datawas insufficient at the beginning, we created a model which simulates the routine (normal) or non-routine (abnormal) day, according to the variance of frailty indexes over a six-month period