Underwater acoustic and optical data fusion has been developed in recent decades. Matching of underwater acoustic and optical images is a fundamental and critical problem in underwater exploration because it usually acts as the key step in many applications, such as target detection, ocean observation, and joint positioning. In this study, a method of matching the same underwater object in acoustic and optical images was designed, consisting of two steps. First, an enhancement step is used to enhance the images and ensure the accuracy of the matching results based on iterative processing and estimate similarity. The acoustic and optical images are first pre-processed with the aim of eliminating the influence of contrast degradation, contour blur, and image noise. A method for image enhancement was designed based on iterative processing. In addition, a new similarity estimation method for acoustic and optical images is also proposed to provide the enhancement effect. Second, a matching step is used to accurately find the corresponding object in the acoustic images that appears in the underwater optical images. In the matching process, a correlation filter is applied to determine the correlation for matching between images. Due to the differences of angle and imaging principle between underwater optical and acoustic images, there may be major differences of size between two images of the same object. In order to eliminate the effect of these differences, we introduce the Gaussian scale-space, which is fused with multi-scale detection to determine the matching results. Therefore, the algorithm is insensitive to scale differences. Extensive experiments demonstrate the effectiveness and accuracy of our proposed method in matching acoustic and optical images.
With the cost reduction of underwater sensor network nodes and the increasing demand for underwater detection and monitoring, near-land areas, shallow water areas, lakes and rivers have gradually tended to densely arranged sensor nodes. In order to achieve real-time monitoring, most nodes now have visual sensors instead of acoustic sensors to collect and analyze optical images, mainly because cameras might be more advantageous when it comes to dense underwater sensor networks. In this article, image enhancement, saliency detection, calibration and refraction model calculation are performed on the video streams collected by multiple optical cameras to obtain the track of the dynamic target. This study not only innovatively combines the application of AOD-Net’s (all-in-one network) image defogging algorithm with underwater image enhancement, but also refers to the BASNet (Boundary-Aware Salient network) network architecture, introducing frame difference results in the input to reduce the interference of static targets. Based on the aforementioned technologies, this paper designs a dynamic target tracking system centered on video stream processing in dense underwater networks. As part of the process, most nodes carried underwater cameras. When the dynamic target could be captured by at least two nodes in the network at the same time, the target position could then be calculated and tracked.
With the increasing demand for underwater resource exploration, three-dimensional (3D) reconstruction technology is important for searching for lost underwater civilizations, underwater shipwrecks, or seabed structures. However, faced with the limitations of underwater unmanned systems in terms of energy, bandwidth, and transmission delay, 3D reconstruction technology based on video streams as direct data will not work well. We propose a terminal image processing strategy to save data transmission time and cost and to obtain 3D scene information as soon as possible. Firstly, we propose an adaptive threshold key frame extraction algorithm based on clustering, which extracts key frames from the video stream as structure from motion (SFM) image sequences. On this basis, we enhance the underwater images with sufficient and insufficient illumination to improve the image quality and obtain a better visual effect in the 3D reconstruction step. Additionally, we choose global SFM to construct the scene and propose a faster rotation averaging method, the least trimmed square rotation averaging (LTS-RA) method, based on the least trimmed squares (LTS) and L1RA methods. It is proven to reduce 19.97% of the time through experiments. Finally, our experiments demonstrate that the dense point cloud saves about 70% of the transmission cost compared to video streaming.