Performance Analysis of The Effect Euler Regression on Complex YOLOv4 Model for Autonomous Driving Applications
Putri Wahyu Herlina, Suryo Adhi Wibowo, Agus Pratondo
Abstract
Autonomous driving is a technology with a driving automation mode where objects will be captured using LiDAR sensors and cameras. Objects captured by LiDAR will be processed into a point cloud representing the bounding box predicted. However, the detection of 3D objects in autonomous driving in real time causes the distribution of the location of the bounding box coordinates of the predicted results to be out of accordance with the predetermined ground truth. Based on these problems, this research designed and implemented an object detection model using Complex YOLOv4 that focuses on exploiting regression parameters. Exploitation of regression parameters is carried out on upper factors, limit angles, and scaling ranges. Based on the results obtained, the highest mAP value is with an upper factor of 0.4, a limit angle of 25 degree, and a scaling range (0.8;1.05) with an mAP value of 37.7% or 4.9% superior to the Complex YOLOv4 model which has an mAP value of 32.8%. In this research obtained the value of precision, recall, and F1-Score for each class shows the car class is dominant of all classes because of the dataset is imbalanced.