Underwater Deep Object Tracking

Abstract

Visual object tracking (VOT) is becoming an increasingly active research area, and more and more tracking algorithms are proposed every year, as tracking plays an important role in a wide range of realistic applications, such as autonomous driving, robotics, video surveillance, and security. Deep learning methods have demonstrated encouraging performance on open-air visual object tracking benchmarks, however, their strength remains unexplored in underwater video sequences. Apart from the challenges of open-air tracking, videos captured in underwater environments pose additional challenges for tracking such as low visibility, poor video quality, distortions in sharpness and contrast, reflections from suspended particles, and non-uniform lighting. Low visibility and distortions of underwater data created by the contrast and clarity of depth, color, and nature of the water are the two main challenges facing underwater object tracking, which makes the tracking difficult to achieve high accuracy and speed. To overcome these two problems, pre-processing of underwater data is needed to address low visibility and distortions and to improve the quality of underwater data for better object tracking by subsequent trackers in complex underwater environments. A deep learning algorithm will be developed to solve the problem of low visibility and distortions in underwater data in our research on underwater deep object tracking. The goal of this research is to track objects efficiently using multiple deep neural network trackers in unpredictable underwater environments with high speed, accuracy and success rates, and then to evaluate the tracking performance among the various trackers and determine the best one.

Date of Award	Dec 2022
Original language	American English
Supervisor	Sajid Javed (Supervisor)