The Detection And Recognition System Implementation Of Underwater Target Based On VxWorks

Underwater target detection and recognition is the key technology of the underwater intelligent robot's perception of target information and independent operation. This paper derived from Harbin Engineering University's Underwater Robot National Defense Key Laboratory Fund projects. Aiming at the influence of illumination condition and water quality on underwater images, detection and recognition algorithms of underwater target based on feature matching were studied. The main study content included three aspects.The advantages and disadvantages of the traditional image segmentation algorithms were analyzed in this paper. Combined the characteristics of underwater target image, made the Canny edge detection and iterative threshold segmentation complementary, the Post-Processing combined segmentation algorithm was realized, which greatly reduced the noise effect on target extraction such as non-uniform illumination. Target detection and extraction reduced the complexity of subsequent recognition algorithm.The adaptability of the widely used algorithms in target recognition was also analyzed in this paper. A feature matching algorithm based on Hough transform and geometrical feature was proposed. Both the Hough transform line matching polygon algorithm and extracting elliptic target's parameters matching algorithm through moment feature combined with MER have preferable anti-noise ability. In certain extent, they have the performance of anti-rotation and anti-scaling.The development platform was based on Senbo PC 104 combined with embedded real time operating system VxWorks, which loaded the image processing components WindML. In Tornado development environment, accurate detection and recognition of four kinds of underwater target were realized in this paper, which were triangular prism, quadrangular prism, spheroid and ellipsoid. The experimental results showed that underwater target detection using Post-Processing combined segmentation and identification algorithm based on Hough transform straight line matching and geometrical feature matching have preferable accuracy and stability performance.