Systematic Review on Various Deep Learning Models for Object Detection in Videos

In recent decades, deep learning (DL)-based frameworks for object detection made significant strides in the field of computer vision. The primary goal of object detection is to identify objects belonging to specific target classes within an image and assign a label to each object according to its precise location. In contrast to conventional, handcrafted object detection modules, DL-based approaches become capable of extracting both spatial and semantic features from images. This article offers a comprehensive review of DL-based object detection frameworks starting from 2014. This review paper is categorized into two distinct approaches: i) the region-based approach and ii) the region-free approach. Additionally, a diverse dataset used for the training of neural networks and various evaluation metrics to assess performance against state-of-the-art results is also listed. Our experimentation involved the utilization of models like Mask-RCNN, YOLOv4, YOLOv5, and YOLOv7 on the COCO dataset. Through our analysis, it becomes evident that YOLOv7 surpasses the other models in accuracy and speed, showcasing superior performance in object detection tasks. Furthermore, we shed light on the prominent applications of DL-based object detection frameworks and provide some forward-looking recommendations for the future.