Optimizing Object Detection Efficiency for Autonomous Vehicles through the Integration of YOLOv4 and EfficientDet Algorithms

Advancements in autonomous vehicles are highly dependent on object detection systems that need to balance accuracy and speed. This paper addresses the comparative analysis of three object detection algorithms such as integrated YOLOv4-EfficientDet model, stand alone YOLOv4, and stand alone EfficientDet for improved efficiency in autonomous vehicles. Yet the integrated YOLOv4-EfficientDet model, which is the result of a merger of real-time capabilities of the YOLOv4 and computational efficiency of the EfficientDet, turns out to be the best one. It attains the best Mean Average Precision (mAP) of 0.85 compared with the other algorithms, which indicates an excellent object detection accuracy. With a slightly decreased inference speed of 45 FPS frames per second, the integrated model shows a good balance between accuracy and speed. However, the standalone YOLOv4, known for its speed of 55 FPS, has a lower mAP value of 0.78. Although it gains a respectable mAP score of 0.82, the Standalone EfficientDet falls behind in inference speed at 35 FPS. Our comparative analysis highlights the utility of the YOLOv4-EfficientDet model in improving the efficiency of object detection for autonomous automobiles. It offers a convincingly accurate solution with a reasonable inference speed. The selection between these algorithms should focus on the unique needs and constraints of the autonomous vehicle system. This study's findings highlight the potential of using integrated models to improve the ability of autonomous vehicles to perceive their environment, making them safe and reliable on the roadways.