Underwater Target Object Detection (Part 1)

 

Underwater Target Object Detection (Part 1)button3

By: Mohd Faid Bin Yahya (PhD Student)

 

 

There are a number of recent computer vision techniques have been applied to detect objects in underwater. Their performances had also been analysed. This article focuses on two underwater target object detection methods which are feature-based approach and template based approach.

a) Feature-based approach

This method analyzes the correspondence between features of detected object from an acquired image to the one from the database. If the similarity between both of the images is above or below a certain threshold, then a decision is made. In order to overcome the invariances in term of translation, rotation, and scaling, a method called scaled-invariant feature transform (SIFT) had been proposed by Lowe (2004).Thealgorithm includes the following four steps:

(i)            scale-space extrema detection,

(ii)           keypoint localization,

(iii)          orientation assignment,

(iv)         keypoint descriptor.

The only drawback of SIFT algorithm is that it is computationally extensive. Instead speeded up robust features (SURF) had been introduced by Bay et al. (2006) which incorporates SIFT algorithm but requires less processing hence faster computational speed.

 

3a apr15

Fig. 1 Feature matching test using SURF in air. (Upper left corners) Target image patches. (Lower portions) Input images. White lines denote the matched pairs of two feature points. White boxes indicate the successful detection of the objects [1].

 

Reference(s):

  1. L. Donghwa, K. Gonyup. K. Donghoon, M. Hyun, and C. Hyun-Taek. (2012). Vision-based object detection and tracking for autonomous navigation of underwater robots. Ocean Engineering48(2012)59-68.