AN IMPROVISED ENSEMBLE CNN ALGORITHM FOR DETECTING VIDEO STREAM IN MULTIMEDIA

Abstract

The only criteria that are used to evaluate the various neural network-based object identification models that are currently in use are the inference times and accuracy levels. The issue is that in order to put these new classes and situations to use in smart cities, we need to train on them in real time. We were not successful in locating any research or comparisons that were centered on the length of time necessary to train these models. As a direct consequence of this, the initial reaction times of these object identification models will consistently be quite slow (maybe in days). As a consequence of this, we believe that models that put an emphasis on the speed of training rather than accuracy alone are in significant demand. Users are able to gather photos for use in training in the present by utilizing concept names in online data collection toolkits; however, these images are iconic and do not have bounding boundaries. Under these conditions, the implementation of semi-supervised or unsupervised models in a variety of smart city applications might be able to contribute to an improvement in the precision of data derived from IoMT. In this study, we categorize the video clips into their appropriate classes using an improved ensemble classification model.

Keywords

• cnn
• ensemble
• video stream