In this thesis, we propose new methods and systems to detect pseudo periodic behavioral patterns of people with physical and mental disabilities. We have defined the infrastructure of a non-intrusive, cost effective and user friendly system to assist patients with behavioral problems which may be due to cognitive impairments due to mental disorders such as Autism Spectrum Disorder (ASD), Traumatic Brain Injuries (TBI) or Alzheimer's disease. This dissertation starts by describing advances in sensors and monitoring systems. The system is structured to monitor motions or motor patterns by obtaining sensor data from an in-home and wearable wireless sensor system, and to give reminders and feedback to users and assist therapists and caregivers. We then present the system's applications and results in detecting and classifying the behavioral patterns in activities of daily living (ADLs). Unlike previous work which issues a reminder like an alarm clock, our system minimally intervenes with the user only when needed by detecting, classifying and monitoring the tasks. The system is flexible and can easily adapt to subject variability with minimal trainings, and the same algorithm can be used to adapt to new ADLs. To better assist cognitively impaired patients, the system detects incompletion and interrupted activities of the subject and issues a reminder/feedback in an intelligent manner. Our system uses three different sensor platforms to monitor and detect abnormal state to better assist the patients with right guidance at the right time. To achieve the goals defined above, development of signal processing methods based on Gaussian Mixture Model (GMM) and Sequential Classifier from Time Domain and Frequency Domain features are discussed. Data fusion to optimally select, combine and manage sensors from different platforms which possess various characteristics and sampling frequencies to collect data is addressed. A key contribution is the selection of a subset of sensors to be monitored and processed at any given time to reduce computation load and limit providing feedback to the patient only when needed. Sensor data fusion methods address how to combine the information obtained from selected sensors in an intelligent for analysis and classification. We explore automatic extraction of features across sensors in the time-frequency plane. We also investigate several behavior recognition strategies for comparison purposes. Algorithm to detect novel patterns is proposed. The novel pattern detection algorithm to find patterns unknown to the system at the time of training is critical as behavioral patterns change or new patterns are developed. An on-line unsupervised learning method to detect and track novel patterns by analyzing features from Higher Order Statistics is also proposed. The proposed algorithm is tested over 60 hours of data collected across 20 subjects and 4 autistic patients with classification accuracy of 94.6%. Finally, a different sensing platform was investigated to enhance the wearability and comfort level of the user for long term monitoring. We showed that using an array of stitched stretch sensors on every-day wear is feasible and demonstrated its potential for activity detection. We also showed that using a combination of different platforms to complement sensing modalities can be beneficial to improving the classification accuracy of the system. We show that the proposed combination of Gaussian Mixture Model with a sequential classifier is efficient and allows potential for real-time application of the activity detection system. This thesis establishes that despite the similarities in the activities it is possible to accurately detect and classify the specific behavioral patterns. The results are compared with the previously developed methods and show that the proposed method can detect the activities with high accuracy and also allows novel event detection to adapt to the behavioral patterns to a user.