Tracking multiple objects under merging, splitting, and occlusion situations is a challenging task for a surveillance system, especially when objects are close together. This is due to ambiguity and uncertainty of visual features from these objects as compared to a situation where tracked objects are isolated. To handle these challenges in multiple object tracking, techniques that apply observations from detected objects have been introduced. Joint probabilistic data association (JPDA) are one of them and calculate the probability of all possible data association between objects and observations. The advantages of JPDA and particle filter can then be combined to obtain better tracking results from human detection algorithm. The detection stage can be performed using Local Binary Pattern (LBP). LBP is a texture descriptor that combines local primitives into a feature histogram. LBP and its extensions outperform existing texture descriptors both with respect to performance and to computational efficiency. It is then suitable for foreground detection or background modelling. Most methods for multiple camera tracking also rely on accurate calibration to associate data from multiple cameras. However, it is often not easy to have an accurate calibration in some real applications due to practical reasons. The inaccurate calibration can then lead to wrong data association of objects between cameras. To handle the data association of objects in multiple cameras under inaccurate ground plane homography, the RFS Bayes filter can be applied. Observations measurements can be modelled from cameras to a random finite set. This random finite set includes the primary measurement from the object, extraneous measurements of the object, and clutter. Experimental results including challenging cases such as occlusions and merging persons will be presented.