Intelligent monitoring of movements and activities in a network of depth sensors

In many multi-sensor and robotic monitoring systems such as smart homes and surveillance applications, detection of the human body, and recognition of its posture/action play an important role. This, as a result, demands the design and development of an efficient deployment system. In this thesis, we propose an intelligent monitoring system to monitor the movements of the subject using multiple depth sensors. Inspired by the notion of swarm robotics, sensing, and minimalism, we have introduced the notation of 1D scans (1 row of the point cloud) which compared with the traditional analysis using a full point cloud model, offers a lower computation and power consumption among the networked sensors and robots. To start, we have shown the effectiveness of 1D scans in sleep posture detection which is an important information in several health-related applications such as apnea prevention and elderly care. Through our study, it is shown that our proposed method offers better performance than that of the joint-based method which uses the extracted human body joint as the features to provide sleep posture estimation. Based on 1D scans we proposed two different change detection algorithms that can be used as a part of a sensor scheduler in a centralized network configuration. The first method relies on the characteristics of the sensor. In the second method, we have improved the change detection performance and usability by using noise characterization to free up the dependency on the sensor type. This initial change detection is then used as a basis for several follow-up tasks such as foreground segmentation, background detection, target detection, and tracking for monitoring tasks. We also propose posture detection methods based on 1D scans and sample points. The results show high accuracy on estimating the main posture of the subject among (laying down, sitting, bending, and standing). We also analyse the details associated with the postures and analyze the optimized number of scans to accomplish such a task. Our main objective is to investigate how the reduced number of training data through a collection of 1D scans of a subject is related to the rate of recognition and how we can use minimum information from distributed sensing to accurately detect the human body and its posture. The proposed monitoring system consists of two stationary sensors and one mobile sensor. The proposed setup can efficiently monitor an environment with an extended field of view to detect the body posture and estimate the detail of its movement. The schedular controls movements of the robot to extend the monitor area which is not within the range of the stationary sensors. This also results in a reduced number of stationary sensors.