Tracking of Underwater Objects with Occlusion Awareness using an Adaptive DEEP SORT and GMM Approach

The underwater ecosystems exhibit unique challenges for recognizing underwater objects beneath the water. When light penetrates far into the water, it experiences diffraction and scattering. As a result, the images captured are hazy and obscured, making interpretation difficult. In underwater conditions, tracking the object of interest from successive frames frequently results in occlusion of objects. An enhanced Deep Simple Online Real-time Tracking (Deep SORT) mechanism is proposed for the tracking of obscured underwater objects. Initially the objects are detected by the Gaussian Mixture Model (GMM). The location of the entities in the underwater sequences is determined using an adaptive deep SORT method and a deep learning technique based on long-short term memory (LSTM). The proposed approach is compared with existing state-of-the-art underwater object tracking schemes, and the tracking of occluded objects are evaluated.

Automated summary

Recognizing underwater objects poses unique challenges due to diffraction and scattering of light, resulting in blurry images and occlusion. This study proposes an enhanced deep learning approach for detecting and tracking occluded underwater objects.