Performance analysis of surveillance video object detection using LUNET algorithm

Object detection algorithms have applications in various fields, including security, healthcare and defense. Because image-based object detection cannot exploit the rich temporal information inherent in video data, we suggest long-range video object pattern detection. Standard video-based object detectors use temporal context information to enhance object detection efficiency. However, object detection in challenging environments has received little attention. This paper proposes an improved You Only Look Once version 2 (YOLOv2) algorithms for object detection in surveillance videos, specifically vehicle detection and recognition. We reduced the number of parameters in the YOLOv2 base network and replaced it with LuNet. In the enhanced model, by using LuNet model for feature extraction to extract the most representative features from the image. LuNet is unique neural network architecture, a traditional and very promising algorithm for solving machine learning problems in video data frames. We perform numerous tests to evaluate the efficiency of the suggested approach, and our method outperforms conventional vehicle detection methods with an average accuracy of 96.41%. The study's findings demonstrate that the suggested technique achieves higher f-measure, precision, and error rate than other approaches.