Improvement Object Detection Algorithm Based on YoloV5 with BottleneckCSP
Aria Hendrawan, Rahmat Gernowo, Oky Dwi Nurhayati, Budi Warsito, Adi Wibowo
Abstract
Detecting objects using deep learning technology has the advantage of getting good accuracy. The accuracy obtained depends on the processing time of using deep learning technology. One object detection algorithm is called You Only Look Once (YOLO), which currently has its fifth version or Yolov5. This paper proposes the real-time object detection algorithm with a video dataset recorded on the highway using Yolov5. The increase of YOLOv5 started by adding augmentation data mosaic by the size of 480x480. We practiced the YOLOV5 - BottleNeckCSP model to detect objects and then got the object information divided into six classes. The results of using mosaic data augmentation are [email protected] of 0.984, [email protected] of 0.696 by the precision value of 0.95, and a recall value of 0.98. Our research framework can be applied effectively to improve the performance of object detection algorithms.