Research And Application Of Log Volume Measuring System Based On Quasi-circular Object Recognition

In recent years, along with the development of image processing, patternrecognition and computer technology, computer vision has entered a new and steadyera, and has been widely used in forestry, industry, transportation, medicine,information technology, as well as daily life. The automatic measurement of logvolume, as an important application aspect of computer vision, has become a new hotspot of research.The dissertation analyzes the current research of the measurement technology oflog volume at home and abroad comprehensively, for the main bottlenecks of logvolume measuring system, deeply analyzes shortcomings of some edge detection andquasi-circular contour extraction algorithms, puts forward relevant improvementstrategies from theoretical point, constructs structural model of log volume measuringsystem based on quasi-circular object recognition from engineering point, designsquasi-circular object recognition and partition optimization algorithms from systemimplementation.The dissertation focuses on a few key questions of the measuring system:1. In the image pre-processing module, the dissertation proposes a optimizationalgorithm based on the traditional gradient-weighted smoothing algorithm, reducesthe noise effectively and prevents from the distortion of the image smoothing.2. According to section characteristics of log, combining with wavelet transformmulti-scale features, it designers an improved edge detection algorithm based onmulti-scale wavelet transform. The algorithm optimizes the image quality of thelogs, retains the more complete the edge of logs and better edge continuity.3. Based on the module of the weighted threshold algorithm, proposing anoptimal watershed algorithm based on weighted threshold. It suppressesover-segmentation effectively and retains the better target contour information.4. Presenting a strategy for optimizing the image contour extraction. For theshortcomings of current typical snake algorithms, it takes full account of the snakealgorithm problem of converging to a local extreme easily, uses coordinatetransformation and dynamic programming method, so that snake can search energyfunctional minimum throughout the region.5. For non-round or incomplete log images, take human-computer interactionmeans to fit circle.