LSTM-Based Visibility Detection for Airport Images in Time Series

Atmospheric visibility detection is essential for assuring safety in traffic and aviation. There have been some methods that can indirectly detect visibility by analyzing the images. However, most of these methods ignore the continuous information in the video while getting independent images from the video. It’s also difficult work to defining image features that reflect visibility accurately and comprehensively. To address these issues, this paper proposes a visibility detection method based on image-feature information fusion LSTM (Long Short-Term Memory). For reducing the complexity of calculation and improve the effect of model training, the key image feature values were extracted into a vector, then the input matrix of the LSTM was composed by these vectors in time series. Experiments demonstrated that our method can focus on the key image features and make full use of continuous information in the video, and thus achieve a satisfying performance invisibility detection.