Multi Object Tracking Using Parallel Processing

Tracking or following a single object is not a complex scenario however, suddenly it becomes more complex if there are multiple objects in the close vicinity. It becomes even more difficult if these objects instead of moving in straight line starts maneuvering. Now what are the challenges in multiple object tracking: correct object modeling, prediction, data association and finally error correction. There are two types of models describing an objects path: straight line motion path and a maneuvering object path model. Based on the path model one can predicts objects next possible position, the predicted position is used to make possible data association and then correction is made to get the actual object position. Data association has two approaches firstly, deterministic technique and secondly probabilistic technique. What is described above shows the inherent challenges in building such a system, however, when it comes to implementation of such complex system, computational power and memory are in the highest demand. Therefore, looking in totality one could use more than one processing unit to handle computation and memory demand. This presentation proposed a parallel processing architecture to build an efficient object tracking system and analyze multiple object tracking with in-depth details of the challenges considering various scenarios.