As the elderly world population increases, caregivers are switching to remote care and monitoring solutions to enable their patients to live autonomously at home for as long as possible. Such services are based on detecting and recognizing Activities of Daily Living (ADL) by using diverse types of sensors at the elder's home that transmit activity data to a remote backend for processing. However, these have a limited reach due to high costs, mostly due to elevated prices of the hardware required, as well as the installation costs. Meanwhile, Smart Home technologies are reaching more homes every year. These are based on identical sets of sensors as those prescribed for remote health monitoring systems. Extensive research has taken place in recognizing ADL given the sensor signals. However, these methodologies require prescribed sensors, the floor plan of the house and the location of the sensors. In this thesis, we first propose a generic IoT interface that enables the backend to connect to arbitrary sensors, such as those typically found in Smart Homes. Regardless of the brand and communication protocol they use, the sensor data is transmitted to a backend where it can be processed by the ADL recognition algorithms and other telemonitoring services. We evaluate our design in the Philips HomeLab environment where generic sensors are installed. The system correctly transmits the activity detected by the sensors to a remote cloud service. Then, we propose a methodology based on Multidimensional Scaling to estimate the location and distribution of sensors based entirely on monitored sensor activity. Finally, we propose a methodology combining heuristics and Support Vector Machines to classify and label the sensors by the type of room in which they are placed, such as Kitchen, Bedroom or Bathroom. We train and test our model based on 4 different houses where volunteers had sensors monitoring their activity for at least 80 days. Our proposed methodology correctly identifies 5 different key locations and fully characterizes the interior of the house. Our heuristic classifier achieves a sensitivity of 0.84 and specificity of 0.98. The SVM classifier achieves a precision of 0.78 and recall of 0.78. This results in an automated method of generating the configuration data needed by the fore mentioned ADL recognition algorithms. The experiment demonstrates the feasibility of using the IoT paradigm for remote health monitoring systems by reusing existing sensors found in smart homes, reducing burden and costs which can potentially have an impact on a broader adoption of these kind of monitoring services.