Automatic Hybrid-based Fish Detection with Background Videos
Vrushali Pagire and Anuradha Phadke
Significant advances in underwater video surveillance systems make autonomous item recognition and tracking an important and demanding task. Due to a lack of light, the direction of moving objects, seafloor features, background flexibility of marine florae, and image variation due to the form and texture of different fish species, background removal in underwater fish detection becomes increasingly challenging. A novel hybrid model was designed and applied in this study for fish detection in underwater settings. For object detection, the LSBP method employs a novel attribute extractor named Multi Frame Triplet Pattern (MFTP). MFTP encrypts the layout of the area in the background using three successive frames and takes into account the local variances in the intensity of eight neighborhoods. The hybrid technique makes use of the MoG2 and LSBP algorithms. The suggested system’s performance is assessed using the Fish4Knowledge dataset. The accuracy, recall, and F-measure of the proposed hybrid technique are 0.9966, 0.611, and 0.7949, respectively.
Keywords: Fish Detection, LSBP, Moving Object Detection, MOG2, Underwater Object Detection