object tracking using particle filters and deep convolutional network

There is a useful method for quick and efficient tracking of multiple objects called simple online and real-time tracking (sort). by adding visual information, sort algorithm performance can be improved. the number of identity switches can be minimized by this. a deep network that is offline on a wide data set of qualified pedestrians has been used since the main structure of the algorithm has a lot of computational complexity. in order to extract more and higher quality visual information that can assist the object recognition algorithm, the focus of this article is on the design of this deep network. to enhance data association efficiency, the paper also used a particle filter instead of a kalman filter. on two standard datasets, mot16 and mot17, we checked our proposed method and compared its performance with other available methods. the results indicate that, relative to the current methods in this area, the tracking accuracy (52.2) on the mot17 dataset is increased. experimental assessment demonstrates that in dynamic settings, our proposed architecture increases the number of identity switches and preferably tracks goals.