Design and Implementation of an Underwater Target Tracking System

This page contains information about the Underwater Target Tracking System which was developed with the cooperation of me and my dear colleague Daniele in the University of Sapienza.

There are many different algorithms to implement a tracking system and each algorithm has a different performance in different conditions. However, in under water environment there are several issues that might raise problems in target tracking systems and hence they need to be addressed carefully. Among these prablems the following ones can be high lighted:

• Unclear water: Muggy water creates a blured, and foggy environment where it is hard to find the target. In such a condition the features of the target is blured and mixed with the environment's features.


• Air Bubbles: The air bubbles can be a very distruptive sources of the noise. Devices or creature that generate a lot of air bubbles around themselves can hide themselves in the back of those noisy scenes. Also, the bubbles might distract the tracking algorithms easily and lead them to lose the target.


• Disguising Targets: Some of the object in under water domain might try to hide themselves by exploiting a similar color and/or textture to their background/foreground. A robust tracking algorithm, must be able to have an efficient feature extraction properties to avoid this type of problems.


• Light Reflections: The moving water surface might lead to a moving light in underwater domain. This variant lighting can change the features of the targets easily on one hand. On the other hand the light itself might be considered as a target if the algorithms fail to extract a proper feature set.

As mentioned, there are several algorithms for target tracking that among them Boosting, MIL, CSRT, KCF, MedianFLow and MOSSE can be highlighted as the most well known algorithms. In this project we proposed a new technique called Automatic Threshold Optimizer and implemented a Tracking Algorithm. We also mixed our algorithm with other algorithms that are mentioned. The results show that our proposed algorithm and also its mixture variants with other algorithms, create a more accurate tracking system.

In below, some of the tracking results of this algorithm has been represented: