Certain investigations of enhancing moving object detection in data streams using machine learning models

Abstract

Real-time analysis of video surveillance is an intriguing and challenging application. Comparatively speaking, it is challenging to identify moving objects. Residual Network 18 (RESNET18) and You Only Look Once (YOLO) are two deep learning-based methods for detecting moving objects, although RESNET18 takes more time to analyze while YOLO misses small objects. To detect moving objects, this study compared regional convolutional neural networks to conventional machine learning techniques. newlineTraditional machine learning algorithms like Support Vector Machine to Deep learning techniques like YOLO were analyzed for moving object detection. It is also observed that the tracking and detection are based on individual object-oriented detection. However, for effective surveillance, multiple-moving object detection is required for observing suspicious activities. It is also observed that those techniques required higher processing time for analyzing larger datasets. Based on this, in this research work, multiple object detection methods are proposed using artificial intelligence to achieve both higher accuracy and reduced computational time. newlineTo recognize multiple moving objects like walkers, vehicles, and trucks, the suggested system first uses conventional methods such as feature extraction utilizing Speeded up Robust Features (SF) and Kernel Support vector machine (K-SVM) model. This method used SF characteristics to reach a greater accuracy of 90%, but its performance in precisely identifying fast-moving objects was minimal. newline newline