Implementation of Novel Object Detection and Tracking System Using Spatio-Temporal Attention-Based FPN with Mobile-SSD

The significant part of computer vision is object recognition and tracking. The object tracking algorithm's objective is segmenting the region of interest of the objects for tracking their position and motions. The demand for automated object tracking from video has grown due to the accessibility of affordable, high-quality cameras and with the rapid advancement of deep learning mechanisms. Yet, there are numerous unique issues with the current object detection and tracking systems, including, background clutter, changes in object appearances due to lighting, limited ability to deal with rapid motion, and the requirement for a large amount of training data. To address these challenges, the submitted research presents new automated and efficient object detection and tracking system supported by an advanced deep learning approach. The key innovation of the research lies in integrating the Spatio-Temporal Attention-based Feature Pyramid Network with the Single Shot MultiBox Detector, enabling improved detection of small-sized objects in images. The required input images will be gathered from the available datasets and further, these images are passed as input to the Spatio-Temporal Attention-based Feature Pyramid Network with a Single Shot MultiBox Detector (STA-FPN-SSD) technique for detecting and tracking the targeted objects. Finally, the developed system is analyzed with the conventional models by employing distinct performance metrics for assessing its supremacy.