Tracks In Infrared Image Recognition Technology Research

This master dissertation mainly deals with the automatic extraction of rail. A priori knowledge of the tracks in the infrared image as a guide, the identification of track is based on the general steps of image understanding. The goal of low-level processing is the extraction of the target area and the edge of the tracks, refine the target area, and eventually get the curve based on the shape and location of the tracks. The mid-level processing is the extraction of primitive line segments to describe the curve. The high-level processing screened curve to extract the rails, based on a priori knowledge. This master dissertation studies different algorithms for image processing in various levels of processing and make some improvement in the corresponding algorithm, and mainly analysis a variety of image segmentation algorithm to extract the region or the edge.The threshold method is used for the regional segmentation of the tracks. In the infrared image, track regions performance of the low gray dark stripes, while there are a large number of high brightness of the background regions and low brightness of background. Thus the possibility of directly using threshold method for track extraction is very small, and the rail itself also has some gray interval. In order to overcome these adverse factors, this master dissertation adopts the following methods to solve the problem:(1)According to the observation, the image information is mainly concentrated in the2/3height of the part, and the above part is brightness of the background area, so the selection of the following2/3as part of object processing.(2)Firstly extract the railway area, and then extract the rail region.(3) Another track searching method is based on template, we use the strategy from coarse to fine. The method is not based on global processing, so it can reduce the effect of background region. Combining the gray features and shape features of track, this method uses a rectangular template to search rail paragraph by paragraph by setting the corresponding similarity measure function, the final track target area is relatively simple. The edge extraction is used for the regional segmentation of the tracks. This master dissertation compares and analysis the effect of a variety of edge extraction algorithm acting on track image. The distribution of the rail track cross-section is seen as a certain width of the roof edge, so uses an improved Duda operator to extract more complete centerline of the tracks, while non-maxima suppression and dual-threshold method is used to determine the edge. Tracks centerline extraction is more complete, and the spurious edge rarely.In the middle-level processing, fracture line is connected, and Freeman chain code is used to describe primitive line segments. In the high level processing, these characteristics of the primitive line segments is used for the recognition of rail. Eventually the number of extracting rail segments is seen as the basic unit of measurement to evaluate the track detection performance. Through experimental analysis, using Duda operator can get better result of track extraction, recognition accuracy in the high.