A Real-time Underwater Object Detection Algorithm for Multi-beam Forward Looking Sonar

Producing high-resolution underwater imagery in range of visibility conditions is a capability of critical demand for a number of applications. New generation of forward-scan acoustic video cameras, becoming available for commercial applications in recent years, produce images with considerably more target details than optical systems in turbid waters. Previous computer processing of sonar imagery has dominantly involved target segmentation, classification and recognition by exploiting 2D visual cues from texture, or object/shadow shape in a single frame. Processing of video is becoming more and more important because of various applications that involve target tracking, object identification in search and inspection, self localization and mapping, among many other applications. This paper addresses the image registration problem for acoustic video, and the preprocessing steps to be applied to the raw video from a DID-SON acoustic camera for image calibration, filtering and enhancement to achieve reliable results.

On processing and registration of forward-scan acoustic video imagery

Among autonomous surface vehicles, sailing robotics could be a promising technology for long-term missions and semipersistent presence in the oceans. Autonomy of such vehicles in terms of energy will be achieved by renewable solar and wind power sources. Autonomy in terms of sailing decision will be achieved by innovative perception and navigation modules. The main contribution of this paper is to propose a complete hardware and software architecture for an autonomous sailing robot. The hardware architecture includes a comprehensive set of sensors and actuators as well as a solar panel and a wind turbine. For obstacle detection, a segmentation is performed on data coming from an omnidirectional camera coupled with an inertial measurement unit and a sonar. For navigation and control of the vehicle, a potential-based reactive path-planning approach is proposed. The specific sailboat kinematic constraints are turned into virtual obstacles to compute a feasible and optimal heading in terms of cost of gybe and tack maneuver as well as safety relative to obstacle danger. Finally, field test experiments are presented to validate the various components of the system.

Toward an Autonomous Sailing Boat

Multiple object detection is challenging yet crucial in computer vision. In This study, owing to the negative effect of noise on multiple object detection, two clustering algorithms are used on both underwater sonar images and three-dimensional point cloud LiDAR data to study and improve the performance result. The outputs from using deep learning methods on both types of data are treated with K-Means clustering and density-based spatial clustering of applications with noise (DBSCAN) algorithms to remove outliers, detect and cluster meaningful data, and improve the result of multiple object detections. Results indicate the potential application of the proposed method in the fields of object detection, autonomous driving system, and so forth.

Multiple Object Detection Based on Clustering and Deep Learning Methods.

A robust obstacle detection and avoidance system is essential for long term autonomy of autonomous underwater vehicles (AUVs). Forward looking sonars are usually used to detect and localize obstacles. However, high amounts of background noise and clutter present in underwater environments makes it difficult to detect obstacles reliably. Moreover, lack of GPS signals in underwater environments leads to poor localization of the AUV. This translates to uncertainty in the position of the obstacle relative to a global frame of reference. We propose an obstacle detection and avoidance algorithm for AUVs which differs from existing techniques in two aspects. First, we use a local occupancy grid that is attached to the body frame of the AUV, and not to the global frame in order to localize the obstacle accurately with respect to the AUV alone. Second, our technique adopts a probabilistic framework which makes use of probabilities of detection and false alarm to deal with the high amounts of noise and clutter present in the sonar data. This local probabilistic occupancy grid is used to extract potential obstacles which are then sent to the command and control (C2) system of the AUV. The C2 system checks for possible collision and carries out an evasive maneuver accordingly. Experiments are carried out to show the viability of the proposed algorithm.

Robust underwater obstacle detection and collision avoidance

Autonomous underwater vehicles (AUV) by nature operate in partially unknown environments. Any obstacle lying in the path of the vehicle is a potential mission-terminating threat. Inclusion of a forward looking sensor would provide valuable information to the AUV. Threat assessment and navigation plans would use this information in order to avoid obstacles. Any such system should meet the requirements of an embedded autonomous system, that is small size and low power consumption. The obstacle detection system is to be integrated in the AUV via a common interface protocol. It is the intent of this paper to explore one possible solution to implementing such an obstacle detection system. The system used in this project is a forward looking sonar (FLS). This sonar system is a commercially available unit modified for performing obstacle detection tasks. With less than 4 W total power consumption this sonar can be integrated in an AUV. The small volume of the system allows easy placement in existing small AUV designs. The sonar control software is implementing in DOS on a PC/104 486 CPU. Filtered decision information is presented to the control logic of the existing AUV through a standard interface type (Lontalk network). A grid occupancy search method is used to detect the closest object in the vehicle's path. The region forward of the FLS is sub-divided into various cells. The cells are filled with the raw intensity data collected from the FLS sensor. For each filled cell, a cell signature is computed. The maximum signature cell is extracted from the grid. This cell contains transformed target information such as, range, bearing to target, and cell signature. The scanning scheme performs a first sweep at a short range for a quick detection of close targets, followed by a second sweep at a medium range. Cell signature definition and cell mapping are the research efforts associated with this paper. Experiments are performed on a moving platform, with the ultimate goal of testing the detection system integrated in a small AUV. This low-cost obstacle detection system can be easily attached to the existing FAU platforms, greatly improving the system safety and reliability of an AUV operation in high threat areas.

K. Nelson

Papers

Obstacle detection by a forward looking sonar integrated in an autonomous underwater vehicle