A Comprehensive Survey on Cutting-Edge Multi-Object Detection Techniques in Real-World Scenes Using Deep Learning

Object detection (OD) and localization are essential tasks in the field of computer vision. Real-world applications namely facial recognition, augmented reality (AR), video surveillance medical imaging, and autonomous vehicle driving requires highly reliable and cost-effective approach to detect and localize objects. Conventional machine learning (ML) approaches namely viola -jones, scale-invariant feature transform (SIFT) and histogram of oriented gradients (HOG) enormously relies on manual procedure of feature representations and often struggled with variations in lighting and occlusion problems. Eventually, deep learning (DL) techniques replaces these traditional methods with capability of essential features learning automatically and handling large-scale datasets more efficiently. This article highlights popularity of DL models namely convolutional neural networks (CNNs), region-based CNN (R-CNN), single shot multi-box detector (SSD) and you only look once (YOLO) that exhibits superior performance than classical methods. Recently, a significant shift from supervised to unsupervised approaches to overcome the reliance on large, annotated datasets have made OD increasingly robust, scalable, and adaptable across various domains. Modern techniques namely FreeSOLO, transformer-based and adversarial-based detectors overcome annotations overhead and improve novel or unknown OD performance. At the end of the research article, anticipated research trends in OD domain are high-lighted.