Moving Object Detection and Classification using Deep Learning Techniques

Abstract

This research study focuses on the critical task of moving object detection in videos in order to improve accuracy and efficiency in identifying dynamic features within a scene. The proposed method combines the accuracy of optical flow estimation specifically; the Farneback method with the YOLO (You Only Look Once) model's strong object-detecting capabilities. YOLO is used to identify objects in video frames, and concurrent optic flow analysis is used to identify motion patterns. An innovative method is presented here that evaluates the motion angle and magnitude of each pixel inside a detected object to obtain precise moving object identification. The system decides whether an object is moving by setting a threshold depending on the percentage of pixels showing noticeable motion. With less false positives and greater accuracy, moving objects can be identified thanks to this adaptive technique. The suggested method’s effectiveness in precisely identifying moving objects in a range of circumstances is demonstrated by experimental findings on a variety of datasets. A complete and effective solution for moving object detection in video streams is provided by the combination of optical flow for motion analysis and YOLO for object detection. The method proposed here has potential applications in video analysis, autonomous systems, and surveillance, where accurate detection of dynamic features is critical.