Research On Human Tracking System Based On Wireless Pyroelectric Sensor Network

Human tracking is the procedure which makes sure whether there exists human body and obtains positions of concerned human targets according to the detection signals in tracking system. The human tracking technology can be widely used in kinds of environment surveillance, security and intelligent assistant systems which relate to person security, emergency rescue, protection of valuables and so on. Therefore, the technology has wonderful science significance and market prospective. This thesis is supported by the National Natural Science Foundation and constructs the whole human tracking system based on wireless pyroelectric sensor network. The system hardware uses the pyroelectric sensor network with the Fresenel lens array to realize human movement detection and the system software includes multi-target tracking algorithm and system calibration algorithm.The design of the human system has to deal with following problems and challenges:(1) The design of the structure of pyroelectric sensor hardware, which contains how to improve the detection resolution and optimize the field of view (FOV) of sensor, and the Fresenel lens array.(2) The CPU capacity of sensor node is limited and the network communication bandwidth is severely restricted.(3) The realization of distributed learning algorithm which is designed for multi-human tracking application and is easy for implement of real system and has low complexity.(4) The establishment of exact and extensible simulation platform of distributed pyroelectric sensor network and system testing platform in real environment. The main work of this thesis in allusion to these problems is:First, the thesis designs the pyroecectric sensor which is suitable for human targets tracking after the research of the pyroecectric theory. Then, the thesis constructs the sensor node and gives the target positioning method and network structure in term of the characters of pyroecectric sensor.Second, the thesis gives the mathematic model of target tracking problem and the corresponding mathematic formula for tracking algorithm. At the same time, the tracking system uses the signal processing technology and the structure characters of pyroelectric sensor network to distill the target detecting results for obtaining event sequence signal which provides the original positioning results with the assistance of geometric parameters of network nodes.Third, the distributed multi-target tracking algorithm is based on the system hardware. The procedure of algorithm design contains two parts, namely the data-to-target association and motion filter. For this system, different targets may blot out others in the FOV of sensors, and this situation causes the blur explanation of sensing data and decreases the precision of tracking system. In order to solve this weakness, the data-to-target association is necessary. The motion filter can deal with the detecting data of different targets, eliminate the noise and forecast the next positions of targets to realize target tracking. The system uses the data-to-target association algorithm JPDA which supposes the target as rigidity one and the Kalman filtering consuming less resource to cope with original target data. Moreover, the paper combines the JPDA with distributed Kalman filter for the distributed environment to construct the whole distributed tracking algorithm which prolongs the life cycle of the whole network.Finally, the thesis designs the system calibration scheme in terms of detecting characters of sensor node. When the human target tracking system works in the real environment, it is affected by the environmental noise and hardware parameters’error. It leads to the increase of tracking error in comparison with the simulation situation. So, the calibration algorithm is needed. This scheme requires the calibrating target moves through the line trajectory and does not need any additional equipment for convenience. The calibration utilizes KL divergence to build objective function and information projection to optimize the function to get new node parameters. In addition, the belief propagation is used to obtain distributed calibration scheme which serves the network and improves the tracking results by optimizing node parameters.Based on the above research work, the novelty of this thesis includes three points:(1) The multi-target human tracking algorithm based on the distributed Kalman filter is developed for the system hardware. In this system, the algorithm is the combination of JPDA and distributed Kalman filter for the lower computation and communication cost and this is because it only use the measure covariance and measure data.(2) The calibration objective function is optimized by the information projection technology. According to the mathematic analysis, the node orientation and node position are convergent for m-projection and e-projection respectively, therefore two-step iterative optimization algorithm is developed. This algorithm takes full advantage of multi-nominal distribution and node hardware structure, and guarantees the convergence during the whole procedure.(3) The distributed calibration algorithm is designed by using Gaussian belief propagation. The belief propagation is an effective method of distributed learning, but it is complicated and thus not very suitable for low-cost wireless sensor networks. Therefore the Gaussian belief propagation is the better choice and it is used to build distributed calibration scheme with the help of local calibration. The Gaussian belief propagation can reduce the cost of computation and communication, and meanwhile realizes the distributed calibration in the network.