Real time object tracking with dynamic fuzzy cognitive maps using deep learning

Fuzzy cognitive maps (FCM) is a soft computation method that simulates the operation of the system by expressing some properties of the system for the modeling of complex systems and their relationships on a graph structure. Fuzzy cognitive maps with classical and static weights, which are not supported by any learning method, have some shortcomings and failures in modeling dynamic systems. This is due to the fact that fuzzy cognitive maps with static weights use the same concept relations at the time of operation and for the all conept values. In the literature, different dynamic FCM recommendations are presented to overcome this problem of FCM. In this thesis, a new dynamic point of view for FCM operation is given. The dynamic FCM used here in provides online weight determination with a deep artificial neural network model. In addition, genetic optimization-based FCM weight determination process was used to establish the data set of the deep artificial neural network used for weight determination. In addition, the developed structure has been tested on different virtual scenarios and the results of the thesis study are given. When the results are examined, the error tolerance and high performance in dynamic operations are noteworthy since the structure developed by the deep learning and weight determination process. After testing the structure on virtual scenarios, different object tracking applications with dynamic fuzzy cognitive maps (DBBH) using deep learning were performed in real time. The first object tracking application mentioned here is the classic single object tracking. The proposed method, which has the least complexity, has successfully implemented the tracking application. Then, when the performance of the proposed method on classical multiple follow-up applications is tested, the results obtained were considered as satisfactory and successful. Last, the DFCM structure using deep learning, is tested on a multi view objtect tracking scenario and this test results also show us our new DFCM structure can track object with high performance in a multiview scenario.