The core objective of object detection in computer vision is widely applicable across diverse fields including robotics, surveillance systems, and autonomous vehicles. The You Only Look Once (YOLO) method attracts significant interest due to its real-time performance and high accuracy, particularly in the era of deep learning, which transforms object detection methodologies. This research evaluates various deep learning models, focusing on the YOLO series, for object detection tasks using the VOC 2012 dataset. YOLOv5 emerges as the most balanced model with mAP of 63.8, while YOLOv6 shows good accuracy but requires high-end hardware. YOLOv7 and YOLOv8 exhibit lower accuracy and slower training times. The hyperparameters for the feature-extraction phase are set based on parameters such as learning rate, optimizer, model size, and momentum. Through hyperparameter tuning, YOLOv5m achieves an impressive mAP of 78.4, surpassing other models. Furthermore, we analyze the model’s performance on a restructured dataset. Particularly noteworthy is the robust performance exhibited by YOLOv5m across various datasets, with the animal dataset achieving the highest mAP of 91.3. These results underscore the efficacy of YOLOv5m in object detection tasks and emphasize the critical role of parameter tuning and dataset customization in enhancing detection accuracy. Experiments with YOLOv5m and YOLOv5l, coupled with ensemble techniques, showcase their effectiveness in detecting objects in complex scenes, highlighting the importance of optimization and ensemble strategies.