Local optical-flow estimation for forward looking imaging sonar data

In the underwater environment the use of optical camera technology is limited by the light attenuation and distortions due to turbidity the water column. High frame rate forward looking sonars have been developed as an alternative, however, when compared to optical systems the image is of a lower resolution and suffers from sound imaging related artefacts. This means that there is a need for registration algorithms that are better able to address these specific difficulties. The registration algorithm presented in this paper, uses local phase information to estimate movement within the image for each pixel. A coarse estimate is achieved by using sparse recovery techniques to select from a dictionary of modelled movements. This estimate is then refined using an adaptive filter. The resultant displacement map can then be used to enhance the image quality, or, as the basis for estimating the sensor movement. The method is used to register an example sonar image and the results shown against full image phase based registration.