Fishery surveys that call for the use of single or multiple underwatercameras have been an emerging technology as a non-extractive mean to estimatethe abundance of fish stocks. Tracking live fish in an open aquatic environmentposts challenges that are different from general pedestrian or vehicle trackingin surveillance applications. In many rough habitats fish are monitored bycameras installed on moving platforms, where tracking is even more challengingdue to inapplicability of background models. In this paper, a novel trackingalgorithm based on the deformable multiple kernels (DMK) is proposed to addressthese challenges. Inspired by the deformable part model (DPM) technique, a setof kernels is defined to represent the holistic object and several parts thatare arranged in a deformable configuration. Color histogram, texture histogramand the histogram of oriented gradients (HOG) are extracted and serve as objectfeatures. Kernel motion is efficiently estimated by the mean-shift algorithm oncolor and texture features to realize tracking. Furthermore, the HOG-featuredeformation costs are adopted as soft constraints on kernel positions tomaintain the part configuration. Experimental results on practical video setfrom underwater moving cameras show the reliable performance of the proposedmethod with much less computational cost comparing with state-of-the-arttechniques.
展开▼