Examining fish behaviour through acoustic tracking is a technique being employed more and more. Typically, research using this method focuses on detections without fully considering the influence of both the physical and acoustic environment. Here we link the aquatic environment of Cumberland Sound with factors influencing the detection effectiveness of fish tracking equipment and found multi-path signal interference to be a major issue while seasonal variabilty had little impact. Cumberland Sound is a remote Arctic embayment, where three species of deep-water fish are currently tracked, that can be considered as two separate layers. Above the 300 m deep sill, the cold Baffin Island Current follows a geostrophic pattern, bending into the sound along the north shore, circulating before leaving along the south shore. The warm deep water is replenished from the recirculated arm of the West Greenland Current occasionally flowing over the sill and down to a stable depth. This influx of water prevents deep water hypoxia, allowing the deep-dwelling fish populations in the sound to thrive. To complement the work done in Cumberland Sound, a year-long study of the underwater soundscape of another Arctic coastal site, Cambridge Bay, Nunavut, was conducted over 2015. Unlike other Arctic locations considered to date, this site was louder when covered in ice with the loudest times occurring in April. Sounds of anthropogenic origin were found to dominate the soundscape with ten times more snowmobile traffic on ice than open water boat traffic. / Graduate

As the largest unexplored area on earth, the underwater world has unlimited at traction to marine scientists. Due to the complexity of the underwater environment and the limitations of human divers, underwater exploration has been facilitated by the use of submarines, Remotely Operated Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs). In recent years, use of autonomous control systems being integrated with visual sensors has increased substantially, especially in marine applications involving guidance of AUVs. In this work, autonomous fish-tracking via AUV with vision servoing control system is studied with the purpose of assisting marine biologists in gathering detailed information about the behaviors, habits, mobility, and local and global distributions of particular fish species. The main goal of this work in this thesis is to develop an AUV sensing system, including both video and auxiliary sonar, which has the ability to carry out visually guided autonomous tracking of a particular species of fish, Large Mouth Bass. A key in enabling fish-tracking involves the development of a vision-processing algorithm to measure the position of the vehicle relative to the fish. It is challenging because of the complex nature of the underwater environment including dynamic and varied lighting conditions, turbulent water, suspended organic particles and various underwater plants and animals, and the deformable body of fish while swimming. These issues cause target fish identification by computer vision processing extremely difficult. In automated fish-tracking work, we provide two valid and efficient segmentation and recognition vision algorithms to identify a fish from the natural underwater environment: one is a feature extraction algorithm based on Gabor filter texture segmentation and a new approach that we call projection curve recognition. It is able to extract the feature on the fish tail and body and successfully describe the fish as two straight line segments. The second algorithm is SIFT based fish recognition algorithm. The SIFT approach introduced by David Lowe in 1999 extracts distinctive invariant features to scaling, illumination, rotation or translation of the image. The reliable keypoints matching in the database of keypoints from target fish is implemented by Best-Bin-First (BBF) algorithm. Clustering keypoints that agree on the possible object with Hough transform are identified as the object fish, reliable recognition is possible with as few as 3 features. Finally, a dynamic recognition process was designed using continuously updated fish model to match and recognize the target fish from a series of video frames. The SIFT Based recognition algorithm is effective and efficient in identifying Large Mouth Bass in a natural cluttering underwater environment. For a monocular camera system, the depth of field is extremely hard to obtain by vision processing. Hence, the system is augmented with a forward-looking digital image micro sonar. With the sonar image processing algorithm, the target fish is recognized. Sonar can not only provide the relative range between the fish and AUV, but also assist in identifying the target. Finally, the relative position and orientation of the fish in the image plane is estimated using an image processing method, transforming the coordinates between camera, sonar and AUV, and applying the estimation algorithm. The results of o off-line data processing taken from a natural Lake environment shows these computer vision algorithms for identifying fish and state estimation are efficient and successful. The proposed system has potential to enable a vision servo control system of AUV to reliably track a target fish in natural underwater environment.

As the largest unexplored area on earth, the underwater world has unlimited at traction to marine scientists. Due to the complexity of the underwater environment and the limitations of human divers, underwater exploration has been facilitated by the use of submarines, Remotely Operated Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs). In recent years, use of autonomous control systems being integrated with visual sensors has increased substantially, especially in marine applications involving guidance of AUVs. In this work, autonomous fish-tracking via AUV with vision servoing control system is studied with the purpose of assisting marine biologists in gathering detailed information about the behaviors, habits, mobility, and local and global distributions of particular fish species. The main goal of this work in this thesis is to develop an AUV sensing system, including both video and auxiliary sonar, which has the ability to carry out visually guided autonomous tracking of a particular species of fish, Large Mouth Bass. A key in enabling fish-tracking involves the development of a vision-processing algorithm to measure the position of the vehicle relative to the fish. It is challenging because of the complex nature of the underwater environment including dynamic and varied lighting conditions, turbulent water, suspended organic particles and various underwater plants and animals, and the deformable body of fish while swimming. These issues cause target fish identification by computer vision processing extremely difficult. In automated fish-tracking work, we provide two valid and efficient segmentation and recognition vision algorithms to identify a fish from the natural underwater environment: one is a feature extraction algorithm based on Gabor filter texture segmentation and a new approach that we call projection curve recognition. It is able to extract the feature on the fish tail and body and successfully describe the fish as two straight line segments. The second algorithm is SIFT based fish recognition algorithm. The SIFT approach introduced by David Lowe in 1999 extracts distinctive invariant features to scaling, illumination, rotation or translation of the image. The reliable keypoints matching in the database of keypoints from target fish is implemented by Best-Bin-First (BBF) algorithm. Clustering keypoints that agree on the possible object with Hough transform are identified as the object fish, reliable recognition is possible with as few as 3 features. Finally, a dynamic recognition process was designed using continuously updated fish model to match and recognize the target fish from a series of video frames. The SIFT Based recognition algorithm is effective and efficient in identifying Large Mouth Bass in a natural cluttering underwater environment. For a monocular camera system, the depth of field is extremely hard to obtain by vision processing. Hence, the system is augmented with a forward-looking digital image micro sonar. With the sonar image processing algorithm, the target fish is recognized. Sonar can not only provide the relative range between the fish and AUV, but also assist in identifying the target. Finally, the relative position and orientation of the fish in the image plane is estimated using an image processing method, transforming the coordinates between camera, sonar and AUV, and applying the estimation algorithm. The results of o off-line data processing taken from a natural Lake environment shows these computer vision algorithms for identifying fish and state estimation are efficient and successful. The proposed system has potential to enable a vision servo control system of AUV to reliably track a target fish in natural underwater environment.

Tracking an individual specimen can be a difficult task especially when one also has to keep track of the environmental factors that affect the tracked specimen’s behavior. The task of tracking these animals becomes impossible when they become submerged in water and their number increases to more than just one. The aquatic species that are being tracked by this project in Lake Pontchartrain and the Gulf of Mexico are: tarpon, scalloped hammerhead, whale shark, tiger shark, yellowfin tuna, spotted seatrout, redfish, and bull shark. We are tracking these fish using acoustic and satellite transmitters. The insertion of transmitters in the fish was handled by the Louisiana Department of Wildlife and Fisheries biologists. The acoustic transmitters were implanted on smaller fish that only swam in Lake Pontchartrain. Due to this, receivers were only implanted at locations across the lake on various types of attachments such as buoys, PVC pipes, and pilings. These receivers were positioned at more than ninety locations in order to maximize the acquisition of detections. These species were tracked in Lake Pontchartrain and the Gulf of Mexico. After this preliminary setup, a constant batch of data was generated on a regular basis and this data was process by the application developed in this project. A Ruby on Rails application was then setup in order to store this data and manipulate it to display an animated track. The application utilizes: Ruby, Rails, HTML, CSS, SQL, JavaScript and multiple third part libraries. Many optimizations were performed in order to ensure reliability and performance when loading a high volume of fish or if a high volume of users were to use the application.