Bed-exit behavior recognition can be the first line of defense to prevent a subsequent fall and injuries, especially for patients with a high fall risk. The techniques adopted to recognize bed-exit behavior include sensor- and vision-based processing. Generally, vision-based techniques can obtain a wide range of activity information to ensure a good recognition performance. Privacy concerns, however, impede the potential use of vision-based techniques and require the monitoring of activities in only a limited region. This paper focuses on behavior analysis using sensor-based techniques to deal with privacy concerns and other practical issues such as environmental cleaning and behavior differentiation between patients and caregivers. A local object tracking (LOT) technique based on an array of multiple reflective infrared (IR) sensors is developed to monitor user activities in a limited region. The proposed IR-based LOT technique utilizes a finite state machine (FSM) to differentiate the bed-exit activities from a caregiver and in-bed user activities. Furthermore, this bed-exit recognition system is realized as a product prototype to examine its performance in a real ward environment. The experimental results show a correct recognition rate of 99% for 26 bedside activities, four of which are caregiver activities, 16 of which are the everyday activities of the in-bed patient, and six of which are bed-exit activities. In a satisfaction survey conducted at a medical institution, 89% of participants (33 caregivers and 22 patients) considered the system to be effective and 90% of them were satisfied with the quality of the bed-exit recognition prototype.