Page 61 - Read Online
P. 61
Sun et al. Intell Robot 2022;2(4):35570 I http://dx.doi.org/10.20517/ir.2022.23 Page 367
A B
Figure 19. Detection result for fishes and seafood. A: detection result with particle filter and Hungarian method, B: detection result
with Fast RCNN method.
Figure 20. Tracking result of an underwater vehicle.
4.3. Intelligent underwater target detection and tracking
For underwater object search and rescue, how to make the designed ROV more intelligent is an important
question, and the key point is intelligent underwater target detection and tracking. For the underwater envi-
ronment, there are basically two target acquisition options: optical images and sonar images. Sonar images
are suitable for searching large areas, while optical images are more suitable for close-range high-precision
operation scenes. Therefore, optical images acquired by underwater cameras have become the main acquisi-
tion means for underwater search and rescue. Some intelligent methods [15,16] are applied for more intelligent
underwater target detection and tracking using images. Different methods are adopted for possible object de-
tection and tracking. Figure 19A shows the detection result with particle filter and the Hungarian algorithm
method for fishes, while Figure 19B presents the detection result with the Fast RCNN method for seafood.
Figure 20 presents the tracking result of an underwater vehicle, while Figure 21 shows the tracking result of
moving fishes.
5. ROV OFFSHORE TEST (TARGET SEARCH AND RESCUE)
To further verify the performance of various indexes of the developed search and rescue ROV, the research
group jointly cooperated with the East China Sea Rescue Bureau to conduct the sea trial experiment. The
test location was selected near the Chicken Bone Reef (122°23’ E, 31°11’ N) in the Yangtze Estuary. This
experimental ROV was carried on the test mother ship Donghai Rescue 118. The lifting and releasing of the
whole ROV were completed by the crew of the Rescue Bureau. The members of our research group were
mainly responsible for the laying and control of ROV cables. Figure 22 presents the search and rescue of ROV