Optimizing YOLOv3 with TensorFlow for Accurate and Efficient Object Detection

Abstract

Object detection is a critical task in computer vision, with applications spanning autonomous driving, surveillance, and robotics. In this study, we implemented and evaluated the YOLOv3 model for real-time object detection. The model was tested on various images, demonstrating its ability to accurately detect and classify multiple objects with high confidence. The results indicate that YOLOv3 achieves a mean Average Precision (mAP) of 55–60% on the COCO dataset, aligning with its original performance benchmarks. Additionally, the model operates at an inference speed of approximately 30 FPS on a Titan X GPU, making it suitable for real-time applications. A comparative analysis with other object detection models, such as Faster RCNN and SSD, highlights the trade-off between speed and accuracy, with YOLOv3 offering a balanced approach. The proposed implementation successfully detects objects in complex environments, validating its robustness and efficiency. Future work could explore enhancements through transfer learn- ing, model pruning, and integration with next-generation YOLO architectures.