| Full text | |
| Author(s): Show less - |
Lima, Felipe B. C. L.
;
Doria, Fernando F.
;
Real, Lucas
;
Oliveira, Vinicius R. G.
;
Takimoto, Rogerio Y.
;
Sato, Andre K.
;
Najafabadi, Hossein R.
;
Tsuzuki, Fabio S. G.
;
Tsuzuki, Marcos S. G.
;
Tsuzuki, MDG
;
Pessoa, MAD
Total Authors: 11
|
| Document type: | Journal article |
| Source: | 2021 14TH IEEE INTERNATIONAL CONFERENCE ON INDUSTRY APPLICATIONS (INDUSCON); v. N/A, p. 5-pg., 2021-01-01. |
| Abstract | |
The 3D reconstruction and automatic statistical analysis of sporting events is an area of increasing demand and innovation. In this work, it is proposed the use of two different CNNs (AlexNet and LeNet) with different activation functions for the classification of soccer field view. The images will be classified as viewing the right goal, the center field, and the left goal. The classification of the soccer field view facilitates the creation of an automated algorithm for creating correspondences of relevant points. This activity is the most difficult to be automated. As it is very difficult to obtain real images for training the CNNs, it is proposed a method for creating synthetic soccer field views. Three CNNs presented good results and high accuracy. (AU) | |
| FAPESP's process: | 20/02538-1 - Player detection and ball tracking in soccer game scenes |
| Grantee: | Fernando Ferreira Doria |
| Support Opportunities: | Scholarships in Brazil - Scientific Initiation |
| FAPESP's process: | 19/03453-2 - MEMS-based inertial sensor for smart cities |
| Grantee: | Hossein Rostami Najafabadi |
| Support Opportunities: | Scholarships in Brazil - Doctorate |