<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>In recent years, as one of the important technical tasks in the field of deep learning, object detection has broad prospects and applications in the field of road obstacle detection. However, in the real driving scene, there are many obstacles, serious occlusion, overlap and other problems, so that the existing obstacle detection algorithm can not effectively detect the obstacles on the road, so it can not guarantee the driving safety. In order to solve the above problems, this paper improves on the basis of Yolo V4 algorithm. Firstly, kmeans + + clustering is used to generate a priori box suitable for the data set to enhance the scale adaptability; Then, the ciou is used as the loss function of coordinate prediction to evaluate the coincidence degree of prediction frame and truth value frame more reasonably. Finally, a suitable target detection data set is constructed by preprocessing the public data set cityccaps. The experimental results show that the improved algorithm can achieve more than 90% accuracy for obstacles with large number of targets in the training set. Compared with the original Yolo V4, the average detection accuracy of the improved algorithm is improved by 2.03%.</p>
</abstract>

In recent years, as one of the important technical tasks in the field of deep learning, object detection has broad prospects and applications in the field of road obstacle detection. However, in the real driving scene, there are many obstacles, serious occlusion, overlap and other problems, so that the existing obstacle detection algorithm can not effectively detect the obstacles on the road, so it can not guarantee the driving safety. In order to solve the above problems, this paper improves on the basis of Yolo V4 algorithm. Firstly, kmeans + + clustering is used to generate a priori box suitable for the data set to enhance the scale adaptability; Then, the ciou is used as the loss function of coordinate prediction to evaluate the coincidence degree of prediction frame and truth value frame more reasonably. Finally, a suitable target detection data set is constructed by preprocessing the public data set cityccaps. The experimental results show that the improved algorithm can achieve more than 90% accuracy for obstacles with large number of targets in the training set. Compared with the original Yolo V4, the average detection accuracy of the improved algorithm is improved by 2.03%.

Road Obstacle Object Detection Based on Improved YOLO V4

International Journal of Advanced Network, Monitoring and Controls

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

{"article-title":"Road Obstacle Object Detection Based on Improved YOLO V4"}

In recent years, as one of the important technical tasks in the field of deep learning, object detection has broad prospects and applications in the field of...

AP(%)	Car	Bus	Person	Motorbike	Bicycle	mAP(%)
yolo v4	0.98	0.93	0.92	0.81	0.51	82.95
Improved yolo v4	0.99	0.93	0.92	0.81	0.58	84.98

Hardware environment	processorGraphics card	Intel(R) XEON W-2133Nvidia TITAN XP 12G
Hardware environment	Software Environment	operating system	Ubuntu 16.04
Deep learning framework		Tensorflow-gpu
Programming language		Python
translater		Pycharm2019.1

Parameter	Value	Parameter	Value
LEARN_RATE_INIT	1e-4	MOVING_AVE_DECAY	0.9999
LEARN_RATE_END	1e-6	STAGE_EPOCHS	100

Road Obstacle Object Detection Based on Improved YOLO V4

Publicado en línea: 22 feb 2021

Páginas: 18 - 25

DOI: https://doi.org/10.21307/ijanmc-2021-023

Palabras clave
YOLO v4 Algorithm, Obstacle, Object Detection, Loss Function

© 2021 Xiao Zuo et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Figure 9.

Figure 10.

Figure 11.

Comparison of test results

Environment configuration

Main network parameter values

Road Obstacle Object Detection Based on Improved YOLO V4

Publicado en línea: 22 feb 2021

Páginas: 18 - 25

DOI: https://doi.org/10.21307/ijanmc-2021-023

Palabras claveYOLO v4 Algorithm, Obstacle, Object Detection, Loss Function

© 2021 Xiao Zuo et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Figure 9.

Figure 10.

Figure 11.

Comparison of test results

Environment configuration

Main network parameter values

Palabras clave
YOLO v4 Algorithm, Obstacle, Object Detection, Loss Function