Comparison of machine vision based methods for online in situ oil seep detection and quantification

Bubble detection and quantification is of high relevance for the observation of gas and fluid seeps within the marine environment, e.g. oil leakages or methane seeps. Image sequences using frontal illumination can be used to address this need if robust algorithms are provided for segmentation and volume estimation. The presented work suggests and successfully investigates the application of a segmentation strategy based on the optical flow concept using the Horn Schunck algorithm for laboratory conditions and existing deep-sea video sequences. Volume calculation is performed by two alternative approaches, namely the elliptical best fit and the volume integration method, and compared for a set of rigid bubble replicas. Where both methods show a significant over- respectively underestimation of the total volume, a combination of both approaches proves to be complementary and less error-prone.