Pages: 106-108
Shra Fatima, Maliha Fatima, Wafa Zaidi
In this research, the process of implementing deep learning models for real time object detection using YOLOv4 and SSD are explored. The models were trained and evaluated with dataset from publicly available datasets, transfer learning, preprocessing techniques, metrics like mean Average Precision (mAP) and Intersection over Union (IoU). In the case of YOLOv4 the detection accuracy was better than and the speed was faster than SSD, which was optimal for simpler, resource constrained environments. The models were confirmed to be practical via real time testing with webcam and performance was robust under different conditions. The study also explains the pros and cons of each model and makes recommendations on further improvement of efficiency in surveillance, healthcare, and smart systems such as object tracking, edge deployment, and adversarial robustness.
Object Detection, Deep Learning, YOLOv4, SSD, Real-Time Inference, mAP, IoU, Computer Vision
© kvscsjournal.org . All Rights Reserved.