The real-time reliable detection of the horizon line on high-resolution maritime images for unmanned surface-vehicle

Coarse-to-Fine Evolutionary Method for Fast Horizon Detection in Maritime Images

“…Our study extends the previous method [18] for improving accuracy by adding the factor of global feature to the optimization criterion. Consequently, the accuracy of the proposed method was improved in frames, whose sea color drastically changes.…”

“…Therefore, in our previous study [18], we proposed a novel method that optimizes the parameters of the horizon line. For the fast and accurate detection of the horizon, we considered several improvements.…”

“…Second, instead of considering all pixels of an image for each combination of the horizon line parameters, such as in [10], we defined the local features of the horizon line using a vanishing line characteristic. The result of [18] shows that utilization of the local features of the horizon for optimization criteria can reduce the processing time and can increase an accuracy. However, this method has limitation in certain scenarios such as for mostly occluded horizon and drastic changes in the color of the sea.…”

“…Thus, these methods require a significant amount of time to achieve an accurate detection of the horizon line. In our previous work [18], we determined the local feature of the horizon line, which was used in the fitness function. As a result, the processing speed and accuracy of horizon detection was improved.…”

DOES: A Deep Learning-Based Approach to Estimate Roll and Pitch at Sea

“…The final loss L f inal is then obtained as a simple addition of the aforementioned quantities, as shown in Eq. 3. The GT roll and pitch values have undergone a prior normalization process, which subtracts to each of them the mean and divides by the variance, both calculated over the entire dataset.…”