Tracking the real-time position of an ocean sensor/buoy-like cylindrical target using a depth sensing camera and a computational edge device

Abstract

Positioning and tracking ocean sensor nodes and buoys are very tedious due to ocean currents and periodic cyclones in oceans. These sensor nodes are predominant in present days because these ocean sensors help researchers measure the marine pollution caused by plastics and petroleum. Identifying and extracting data from the sensor nodes and buoys allows us to alleviate adverse climatic impacts and protect the economy, tourism, and fishing industries. Researchers currently employ sonars, both mono cameras and stereo cameras, to visualize aquatic life, coral reefs, and marine waste. In this work, we aim to localize and position a customized cylindrical-shaped sensor-like object using the new generation Intel depth sense camera D455, offering a novel way of determining the distance of underwater objects from the camera. Furthermore, we utilized a new generation NVIDIA AGX Xavier AI computer-aided device to actively track the cylindrical-shaped object in real time. Various positions of the target are assessed, and experiments are conducted to confirm the accurate positioning of the target. It has been confirmed through experimentation that we successfully identified the target up to a distance of 3.7 meters with a good target profile in a laboratory environment. Furthermore, real-time experiments are carried out in a pool using an AI Edge system. The average inference time for each frame obtained from the AI Edge system was 441.3 milliseconds. Also, the accuracy of target detection in video frames reached a maximum of 97%, providing validation for the identified targets.