Improved Small Object Detection for Road Driving based on YOLO-R

With the popularization of self-driving cars, more and more researches have been done on road object detection. However, many challenges remain to be resolved, such as the detection accuracy of small objects in the long distance. Therefore, we propose an algorithm based on YOLO-R to improve the dete...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE International Conference on Consumer Electronics-China (Online) s. 279 - 280
Hlavní autoři: Huang, Yu-Fang, Liu, Tsung-Jung, Liu, Kuan-Hsien
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 06.07.2022
Témata:
Autonomous automobiles
Consumer electronics
Object detection
Roads
Stochastic processes
Training
ISSN:2575-8284
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:With the popularization of self-driving cars, more and more researches have been done on road object detection. However, many challenges remain to be resolved, such as the detection accuracy of small objects in the long distance. Therefore, we propose an algorithm based on YOLO-R to improve the detection accuracy to deal with the actual situation in this field. First, we set some conditions and propose some methods to balance the problem of extremely unbalanced size among each target label. Secondly, the Mish activation function is selected for training. Finally, we use the stochastic gradient descent (SGD) method to ensure that the best global solution can be obtained, and experiments on the BDD100k dataset show that our method has better results than other models in this dataset.
ISSN:2575-8284
DOI:10.1109/ICCE-Taiwan55306.2022.9869118