Ship detection based on spatio-temporal features

The paper proposes the ship detection method based on Spatio-temporal Histograms of Oriented Gradients (STHOG) feature and Support Vector Machine (SVM). STHOG feature, which is the extension version of HOG feature, enables extract spatial and temporal features of an object. The ship detector based on HOG feature can wrongly detect the similar shape objects with ships. On the other hand, the ship detector based on STHOG feature can identify them successfully by utilizing temporal feature of an object. To extract temporal feature of an object, image registration is implemented and an image displacement by camera motion is corrected. Due to high dimensionality of STHOG feature, it requires high computational cost to scan entire image and find ship regions. Principal Component Analysis (PCA) is applied to STHOG feature to compress the dimension. In the computer simulations, the ship detection performance of the proposed method was evaluated. From the simulation results, our proposed method exhibited better results than ship detector based on PCA+HOG feature.