Automated underwater 3D sonar image recognition has great potential to simplify many underwater tasks. For the recognition, sonar modeling is necessary to recognize 3D object images. We proposed a 3D sonar model that can predict what an object is based on recognizing similarities to objects that preexist in a data base. For the modeling, the sonar's displaying mechanism and characteristics of 3D sonar image are studied. Due to the nature of acoustics, a sonar image is not necessarily always an accurate depiction of an object. The proposed model enables mapping of a 3D world onto a 2D sonar screen. This modeling framework enables the implementation of various optical vision techniques for recognition. Recognition experiments were conducted to evaluate the model's accuracy.


Underwater, the image sonar and the optical camera are both feasible sensing methods for object recognition. They are essential to carrying out tasks like object finding, environmental monitoring, and underwater structure installation and maintenance. The optical camera provides the highest underwater image resolution, but has restricted applications due to its limited visibility (Sisman, 1982). Especially since many of the tasks take place in shallow water with very short visibility distances (less than 1m). The image sonar is also a feasible method for recognition. Recently, the quality of the sonar's images has improved dramatically. The latest 3D image sonar such as DIDSON (Dual Frequency Identification SONar: Belcher, 2002; DIDSON website; Kim, 2005; Negahdaripour, 2005) provides high resolution 3D images. Automatic sonar image recognition has great potential in many underwater fields like mine detection, and maintenance and safety inspections. However, automated sonar image recognition's displaying mechanism and the very nature of acoustics both present challenges. These difficulties cannot simply be overcome by the optical camera model based approach.

This content is only available via PDF.
You can access this article if you purchase or spend a download.