Convolutional Siamese-RPN++ and Yolo-v3 based Visual Tracking Regression

307 DOI: 10.37398/JSR.2022.660133 Abstract: Visual tracking is an implementation of moving object tracing from deep machine learning methods where system initially set the object and generate a unique identification or pattern for tracking the moving object at each frame of a video. Object tracking is the undertaking of automatically distinguishing objects in a video and deciphering them as a bunch of directions with high accuracy. This paper intended to propose a SiamRPN network which has been considered as offline network with having very large dataset. In this network there are so many sub networks are available to extract the features along with regression and classification. Here the Siamese-RPN++ has been reconciled with Yolo-v3 which is an object detection approach that enhances the feature extraction model for better visual tracking analysis. Prior recognition frameworks repurpose the classifiers or localizers to perform feature extraction. It applies the model to an image at various areas even while object scaling. System has been tested with various datasets/benchmarks including OTB50 and OTB100 and achieved 91.17 & 89.98 resp. percent of accuracies.