UBC Theses and Dissertations
Detection of fish-food pellets in highly-cluttered underwater images with variable illumination Parsonage, Kevin David
The focus of this research was to develop and test a method using image analysis to detect falling objects in a complex and variable underwater scene. This particular application involved the detection of fish-food pellets in netcage aquaculture systems. The problem was complicated due to the video camera positioning, number of other objects in the scene, and variable and uncontrollable background lighting conditions. The image analysis program was developed and tested using images obtained from industry standard video cameras. Testing conditions were as follows: Food pellet diameter: 2-11 mm. Water visibility: 3.5 - 11 m. - Fish size: 0.025 - 4.8 kg. - Fish stocking density: 0.27 - 20.3 kg/m³. The resulting image analysis program consisted of novel image enhancement and object recognition algorithms and was combined with filtering methods to eliminate false detections. The program was capable of detecting food pellet events providing the following conditions were met: 1) The camera view area was positioned within the sinking path of the food pellets. 2) The camera was positioned with its lens pointed towards the water surface. 3) The camera lens and rigging were clear of debris. 4) At least three food pellets of area 30 pixels or greater were present in the sampled images for 8 consecutive seconds.
Item Citations and Data