Non-extractive fish abundance estimation with the aid of visual analysis hasdrawn increasing attention. Unstable illumination, ubiquitous noise and lowframe rate video capturing in the underwater environment, however, makeconventional tracking methods unreliable. In this paper, we present a multiplefish tracking system for low-contrast and low-frame-rate stereo videos with theuse of a trawl-based underwater camera system. An automatic fish segmentationalgorithm overcomes the low-contrast issues by adopting a histogrambackprojection approach on double local-thresholded images to ensure anaccurate segmentation on the fish shape boundaries. Built upon a reliablefeature-based object matching method, a multiple-target tracking algorithm viaa modified Viterbi data association is proposed to overcome the poor motioncontinuity and frequent entrance/exit of fish targets under low-frame-ratescenarios. In addition, a computationally efficient block-matching approachperforms successful stereo matching, which enables an automatic fish-body tailcompensation to greatly reduce segmentation error and allows for an accuratefish length measurement. Experimental results show that an effective andreliable tracking performance for multiple live fish with underwater stereocameras is achieved.
展开▼
