Code and data for reproducibility.

Abstract — Flight testing is a mandatory process to ensure safety during normal operations and to evaluate an aircraft during its certification phase. As a test flight may be a high-risk activity that may result in loss of the aircraft or even loss of life, simulation models and real-time monitoring systems are crucial to access the risk and to increase situational awareness and safety. We propose a new detecting and tracking model based on CNN, that uses fiducial markers, called HSMT4FT. It is one of the main components of the Optical Trajectory System (SisTrO) which is responsible for detecting and tracking fiducial markers in external stores, in pylons, and in the wings of an aircraft during Flight Tests. HSMT4FT is a real-time processing model that is used to measure the trajectory in a store separation test and even to assess vibrations and wing deflections. Despite the fact that there are several libraries providing rule-based approaches for detecting predefined markers, this work contributes by devel- oping and evaluating three convolutional neural network (CNN) models for detecting and localizing fiducial markers. We also compared classical methods for corner detection implemented in the OpenCV library and the neural network model executed in the OpenVINO environment. Both the execution time and the precision/accuracy of those methodologies were evaluated. One of the CNN models achieved the highest throughput, smaller RMSE, and highest F1 score among tested and benchmark models. The best model is fast enough to enable real-time applications in embedded systems and will be used for real detecting and tracking in real Flight Tests in the future.

Open Access IEEE Latin America Transactions link with video and graphical abstract

@ARTICLE{fast_tracking_melo,
  author={Melo, Gabriel Adriano and Máximo, Marcos and Castro, Paulo André},
  journal={IEEE Latin America Transactions}, 
  title={High Speed Marker Tracking for Flight Tests}, 
  year={2022},
  volume={20},
  number={10},
  pages={2237-2243},
  doi={10.1109/TLA.2022.9885171},
  url={https://latamt.ieeer9.org/index.php/transactions/article/view/6941}
}

Abstract — Hydroelectric power stations are responsible for renewable energy generation, especially in countries with many rivers such as Brazil. It is very important to have good estimates of the hydrological flow in order to determine whether thermoelectric power plants should begin operation, an event that would increase the costs of electricity and also have a terrible environmental impact. The monthly flow of a river was estimated using two recurrent neural networks techniques: Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU). The results were compared with other articles that had the samestructure and used the same data: the Rio Grande river in the Furnas and Camargos dam.

@Article{IEEE_melo2022,
    author={Adriano de Melo, Gabriel and Sugimoto, Dylan Nakandakari and Tasinaffo, Paulo Marcelo and Moreira Santos, Afonso Henriques and Cunha, Adilson Marques and Vieira Dias, Luiz Alberto},
    journal={IEEE Latin America Transactions},
    title={A new approach to river flow forecasting: LSTM and GRU multivariate models},
    year={2019},
    volume={17},
    number={12},
    pages={1978-1986},
    url={https://latamt.ieeer9.org/index.php/transactions/article/view/2224/352},
    doi={10.1109/TLA.2019.9011542}
}