Project Overview
Recently, the work of our Transparent Ocean PhD Student, Hamidreza Farhadi Tolie, Transparent Ocean Lead, Professor Jinchang Ren and RGU School of Computing, Engineering and Technology's Dr Somasundar Kannan and Dr Nazila Fough was published in the peer-reviewed Conference, IEEE International Workshop on the Metrology for the Sea. The research the team has undertaken is partly supported by the wider SeaSense project which is funded by the Net Zero Technology Centre. The paper, titled ‘Promptable Sonar Image Segmentation for Distance Measurement Using SAM', introduces the utilisation of a recently published Segment-Anything deep learning-based approach with an introduced prompt generation method for SONAR images to first automatically identify target objects and then measure their distance to the sensor.
Challenges
The subsea environment poses significant challenges for robotic vision, including uneven light attenuation, backscattering, floating particles and low-light conditions, all of which substantially degrade underwater images. This degradation affects robotic operations that rely heavily on environmental feedback. Sonar imaging offers a solution by using sound pulses instead of light, capturing distance information based on the return time of the sound waves. However, current distance measurements are typically performed manually by drawing polygons on the sonar images and estimating distances. This process is time-consuming and prone to inaccuracies, particularly when the sensor is in motion, making real-time measurements unreliable.
Recent Developments
To address the challenges and develop a more automated solution, the Transparent Ocean team has recently introduced a prompt-based image segmentation approach tailored for use with affordable single-beam sonar devices. This method identifies potential targets and measures their distances from the sensor. Given that segmentation models require large datasets, and single-beam sonars lack shape information, prompt-based segmentation techniques, such as Segment-Anything, help overcome these limitations by precisely identifying objects even in the presence of noise.
The team’s approach consists of the following key modules:
Prompt Generation: Creating effective prompts to guide the segmentation process.
Object/Target Identification: Detecting and isolating potential targets within the sonar data.
Distance Measurement: Accurately calculating the distance from the sensor to the identified targets.
The proposed approach was successfully validated during the SeaSense project demonstration and delivery. The model was deployed on an embedded system, where it processed acquired data in real time. The processed information was then transmitted to our custom-developed software for visualisation, demonstrating its effectiveness and practical utility.
Data Availability
To support the growth of science and foster collaboration, the team is excited to share the data and resources with the broader research community. Interested parties can access the data related to this work at https://github.com/hfarhaditolie/PSIS-ADM. By making this data available, the team aims to contribute to advancements in the field, encourage reproducibility and inspire further innovation. The team would like to invite researchers, developers and enthusiasts to explore, utilise and build upon their work.
Future Plans
Looking ahead, potential plans involve exploring the integration of single-beam sonar images with stereo vision to enhance the depth perception of underwater environments. By combining the precise distance measurements from sonar with the detailed spatial information from stereo vision, this approach aims to overcome the limitations of individual modalities. Such integration could improve the accuracy and resolution of depth maps, enabling more robust navigation and object detection for underwater robotic systems. Furthermore, this approach may facilitate real-time 3D reconstruction of underwater scenes, advancing applications in marine exploration, environmental monitoring and subsea inspection.
