In these activities, feeding means the ability to feed oneself food after it has been prepared and made available. Therefore, eating and drinking detection is a very important topic for daily life surveillance. Measurement of eating or drinking activities in daily life or continuous recording of these activities at home would provide more reliable diagnosis of disabilities for hospitals or insurance companies. However, eating and drinking detection poses a challenge for the state of the art of the research in activity recognition [4], and few references or systematic methods can be found in the literature.In the daily life surveillance system, if the human activities (such as eating or drinking) can be tracked accurately, the results can help greatly and readily improve the ability of the identification of the whole system.

Therefore, devices that can accurately track the pose of limbs in space are essential components of such a surveillance system.One method of tracking and monitoring activities is via tracking the pose of human limbs in space. The human limb tracking system can be classified as non-vision based and vision-based systems. Non-vision based systems use inertial, mechanical and magnetic sensors etc. to continuously collect movement signals. For example, the Micro-ElectroMechanical Systems (MEMS) inertial and magnetic sensor devices [5, 6, 7, 8] can be used in most circumstances without limitations (i.e. illumination, temperature, or space, etc.) and show better performance in accuracy against mechanical sensors.

The main drawback of using inertial sensors is that accumulating errors (or drift) can become significant after a short period of time. Vision-based systems are widely used in recent Cilengitide years, such as [9, 10, 11, 12]. However, most vision-based approaches to human movement tracking involve intensive computations, such as temporal differencing, background subtraction or occlusion handling. In many cases, once a prior knowledge of an estimation of object kinematics is available, the expensive image detector array appears inefficient and unnecessary.Accelerometry-based activity analysis has been developed fast in recent years. Some prototype systems which aim at monitoring daily activities [13], conducting gait analysis [14], etc. are reported. In our system, the 3D accelerometers are applied to collect raw measurement data of the moving arm and the server computer communicates with the sensor devices via the blue-tooth. The simple hardware structure makes the data acquisition and processing easy.