Multiple human tracking is desirable for many applications ranging from surveillance to gait biometrics, robotics and intelligent space. In recent, the low cost, low power consumption sensors, such as acoustic sensor, thermal sensors, pressure sensors, photonic sensors provide a new way to achieve multiple human tracking compared with the conventional tracking, such as Radar, Sonar and video tracking. Advances in sensor network technologies enable the development of distributed tracking systems. In this thesis, we attempt to explain the multiple human tracking problems with our binary pyroelectric infrared (PIR) sensors. We will cover from the designs of hardware sensor nodes to the problems of sensor node selection and calibration to the framework of binary compressive tracking to the implementation of distributed wireless senor system for tracking. In particular, we will investigate (a) A compressive multiple human tracking system using space encoding / decoding methods. (b) A Binary compressive tracking framework using low density parity check (LDPC) matrix and linear programming for encoding and decoding schemes. (c) A distributed information filter algorithm for tracking. (d) The information gain based sensor selection scheme and NMF based calibration scheme. The major accomplishments of this thesis include the following four aspects: (1) Space encoding and measurement decoding schemes. The space encoding scheme is based on the low density parity check (LDPC) matrix, which converts k-sparse target position vectors into different codewords. The measurement decoding scheme contains linear programming based localization and graphical model based tracking algorithms, which converts codewords into the states of multiple targets. A posterior Cram'er-Rao bound analysis is utilized to achieve the tradeoff between the compression ratio of measurements and the accuracy of the tracking system. (2) Information driven sensors selection scheme. The information gain is used to dynamically adding sensors to maximize the information gain. The sensor selection procedure provides the maximum information gain for the whole sensor system which contains the minimum sensor numbers that can be used for tracking. (3) NMF based sensor calibration scheme. We provide a probability model for sensor calibration. Nonnegtive Matrix Factorization method is used to update the probability model. The calibration of sensor parameters, positions and orientations, then can be computed from the updated probability model. (4) Graphical model based problem illustration. We describe the graphical model for tracking algorithms. Various hidden variables can be added in graphical model. Also, a factor graph of tracking is provided to describe the distribute way of belief propagation.