Detection, tracking, identification and classification of multiple moving objects in sequential digital images (ntits)

Due to the developing technology, very high-resolution digital cameras have become more affordable, easier to access, and the areas of their utilization have become widespread day by day. As sequential digital images with very high resolution contain a vast amount of data, however, processing and revealing useful information in a digital environment creates a need for high processor speed and large computing capacity. Even when high processor speed and large computing capacity get to be provided; depending on the algorithms that have been employed, it is often not possible to detect the properties of high-resolution sequential digital images in real-time, extract useful information, and assess them at the same time period. Real-time detection, tracking, identification, and classification of multiple moving objects in sequential digital images require quite special processors with parallel processing capability and proper algorithms for them. In this thesis, a two-step method based on the biomimetic approach was developed for the Detection, Tracking, Identification, and Classification of Multiple Moving Objects in Sequential Digital Images (NTITS). The areas where the focused moving objects are, get detected and separated from the static objects and the background, at the first stage of the method. It was inspired by the principle of how the eyes of humans, monkeys, and some other advanced creatures work when detecting moving areas. The eyes of these creatures do not see every point at sight at the same resolution. While the focused objects are seen with very high resolutions, the sharpness decreases, and the objects become blurred as one moves away from the focused area. In this way, the brain remains not engaged with the details of other areas outside of the focused area. Instead of allocating resources for processing all the pixels in very high-resolution images, only the areas with moving objects are analyzed with the help of this feature in nature. The optical flow (OF) method is applied to separate all variations of animate, inanimate, inductive, capacitive objects from the surface, background, and other static objects. At the second stage of the method, the properties of moving objects are extracted, and the identification, tracking, and classification process take place. Deep Neural Networks (DNN) are employed in object identification and classification processes. As the codes can be processed eminently faster on Graphics Processing Units (GPUs) compared to Central Processing Units (CPUs), GPUs are preferred in the developed method. The results from the NTITS method demonstrated that the detection, tracking, and classification of multiple moving objects in very high resolution sequential digital images can be completed with a high accuracy rate.