As object detection has a close relationship to both image interpretation and video analysis, it has captivated a lot of interest in research recently. Modern object detection methods are based on superficial trainable structures and handcrafted characteristics. Their performance rapidly gets static because they construct complex ensembles that combine high-level information from object detectors and scene classifiers with multiple low-level image properties. As deep learning advances quickly, more influential tools that can understand deeper, higher-level, semantic aspects are being developed to solve issues with conventional architectures. In terms of network design, training methodology been presented a review of SSD, YOLOv9 and Detectron2 – three deep learning based object detection frameworks. Additionally, experimental studies are offered in order to contrast different approaches and derive some insightful findings. Lastly, a number of worthwhile objectives and directions are offered as a basis for future research in the fields of object detection and pertinent neural network-based learning systems, optimisation function, etc., these models exhibit dissimilar behaviours. The study has.