Stabilized Adaptive Steering for 3D Sonar Microphone Arrays with IMU Sensor Fusion

TL;DR

The paper tackles misalignment in in-air 3D sonar imagery caused by uneven terrain when using fixed beamforming. It introduces electronic acoustic image stabilization by fusing the IMU-measured tilt $\theta_i$ to adapt the DAS steering matrix with $\theta_i^c = -\theta_i$ and by applying calibration-based gain compensation, all implemented on a GPU-accelerated pipeline. Key contributions include pre-generating DAS delay matrices across $-30^{\circ}$ to $30^{\circ}$, real-time adaptive steering, and quantitative improvements in temporal consistency of acoustic images. This approach enables robust autonomous navigation and mapping on variable terrain without additional hardware, reducing imaging artifacts and improving measurement reliability.

Abstract

This paper presents a novel software-based approach to stabilizing the acoustic images for in-air 3D sonars. Due to uneven terrain, traditional static beamforming techniques can be misaligned, causing inaccurate measurements and imaging artifacts. Furthermore, mechanical stabilization can be more costly and prone to failure. We propose using an adaptive conventional beamforming approach by fusing it with real-time IMU data to adjust the sonar array's steering matrix dynamically based on the elevation tilt angle caused by the uneven ground. Additionally, we propose gaining compensation to offset emission energy loss due to the transducer's directivity pattern and validate our approach through various experiments, which show significant improvements in temporal consistency in the acoustic images. We implemented a GPU-accelerated software system that operates in real-time with an average execution time of 210ms, meeting autonomous navigation requirements.

Figures (4)

• Figure 1: The processing pipeline from the raw sensor data of the 3D sonar sensor to acoustic images. The adaptive DAS-beamformer uses the measured tilt angle $\theta_i$ captured by the IMU and gains compensation to stabilize the acoustic image result.
• Figure 2: The embedded Real-Time Imaging Sonar (eRTIS) (a) is shown as a drawing at a tilted elevation angle $\theta_i$. The opposite angle $\theta_{i}^c=-\theta_i$ needs to be used to stabilize this sensor. (b) shows the eRTIS sensor mounted on a Clearpath Husky ruggedized mobile platform for outdoor experiments.
• Figure 3: A single validation measurement with (a) the reference acoustic image at $0^\circ$ elevation compared against a measurement at a tilt angle of $\theta_i=-10.78^\circ$ shown in other images, with numbered reflector groups. (b) Unstabilized shows the loss of reflector groups 3 and 4. (c) Fully stabilized image at an adapted angle $\theta_{i}^c=11^\circ$ with gain compensation (x1.34) restores these reflectors. Groups 1 and 2 appear in both. Group 5 are secondary reflections on the sensor, reflected away at a tilted angle. (d) Difference between reference and fully stabilized images. (e) and (f) show the two techniques separately.
• Figure 4: The validation results of all controlled experiments. Cosine similarity was used as a metric for image comparison between the (un)stabilized (at a certain tilt elevation angle) and reference acoustic images.