The Recognition Algorithm Research Of Traffic Road Targets Based On Remote Sensing Image

The recognitionof traffic target in the based on remote sensing image is one of the key technologies of the automatic recognition technology. In this paper the main research content mainly include traffic hinge of line feature such as all grades of highway, bridges, airports and roads,etc. The detection method of these targets can be realized approximately by the same thought and flow. They are edge detection (bottom-level processing), line segment extraction (middle-level processing), line segment connect, target description and extraction (top-level processing). we adopted the method of block edge extraction to extract the road patterns, which is a breakthrough comparing with the traditional method using line characteristic. This paper mainly works includes:1.We studied algorithms of image segment and edge detection, including improved Canny algorithm and morphology algorithm etc. We mainly discussed Snake algorithm. Aimed at the characteristic of the target image, we improved Auto-Snake algorithm by using a kind of automatic active contour model based on Williams’ Greedy method. The improved model can converge steadily at the contour edge of multi-corner image which has a complex distribution with the gray scale of the target and the background. Experiments indicate that this algorithm can be realized to detect the edges of real blocks with any shapes. We proposed an active contour model based on image fusion, which is used for target detection and tracking. First, we apply wavelet transform on multi-resolution image fusion, obtaining the multi-resolution fusion image. Second, we detect target outline using Snake and Sobel operator in the low frequency image. In other high frequency sub-image, we apply a Snake model based on wavelet. These two kinds of convergence Snake fuse into one piece to get the fusion Snake model under the original resolution fusion image. This kind of multi-resolution active contour fusion model can be widely used for target detection and tracking in military or security fields, which have a complicated background.2.We studied the bridge model, and proposed a new kind of bridge target detection algorithm based on improved Hough transform. Theoretical analysis and experiment results show that, our new algorithm has a strong ability of anti-interference and high detection accuracy, deal running time, ideal computation complexity and high detection accuracy, and can give a complete description of the bridge’s target.It is a simple and efficient algorithm for bridge target recognition. The algorithm used the characteristics of bridge in remote sensing images plenarily. After removing the false spikes by the way of combining the global threshold and local threshold, and interpreting the position relationship between the straight line segments, and integrating the improved algorithm such as the fusion endpoint adjacent to the line segments into the standard Hough transform, we realized a new algorithm for bridge target recognition The random Hough transform method has been discussed in many literatures, and we can do some further improvements in this programme to speed up the running speed of the algorithm.3.We proposed two kinds of method for line segment connection,and demonstrated its validity experimentally. The road model and characteristics of different resolution and different scenarios are discussed. For the linear characteristics of the road, first we studied the extraction of linear element. Then we analyzed the Phase Classification and Extraction Method Based on the principles of the Radon transform of these two most representative elements. Then we proposed two linear element connection methods, a linear element connection method based on knowledge, and another curve improved Snake model based on the curvilinear connection. And then we introduced the verification and post-processing of the road goals, including the removal of isolated points and non-linear segments, smooth road goals treatment. A new algorithm about full FFT was introduced.this algorithm is more reasonable and scientific than old FFT. Though this algorithm the research about correlation match for oil storehouses’ recognition was performed. Experiments indicated this algorithm can detect oil storehouses fast and exact, and this algorithm is more efficient and robust Furthermore the relation between template scale feature and oil storehouses’recognition was indicated, that is the algorithm’s speed is increased faster than direct convolution along with the template scale increasing, and small template can detect large oil storehouses, not vice versa。 The precision of the detection will rise along with template scale reducing, but at the same time the false alert will appear.4. we sorted the road model as several classes, and discussed the road recognition method for each road model. According to the remote sensing images of different scenes and resolution, we classified the road model And we discussed methods of road identification for each road model. For remote sensing images of different resolution, we divided them into two categories, one is a low-resolution image, whose image around ground resolution is about 10 meters; the other is a high resolution image, that image around ground resolution of 1 meters.Aiming at detection of highway in the countryside with low resolution, after understandthe model in this situation deeply, we adopted the algorithm based on RT for the highway detection in low-resolution rural areas; and we improved the road detection algorithm based on directional statistical features, that is we balance the direction of linear feature with extracting line feature. These algorithms are proved by experiments and the former one shows a good result in extracting line feature with a structured shapes and the other one has good effect on extracting highway in the open air with a low resolution We discussed the model of city road in high resolution image. Considering the characters of high-density block in images of this kind, we proposed road detection algorithm based on block detection, which means utilizing block edge detection instead of the road extraction. The road detection is achieved by block edge extraction with the improved Auto-Snake algorithm we proposed above. Additionally, aiming at the artificial object such as Military airfield runway, we present anrecognized example in this situation with combining the previous object recognition flow.In short, the paper has carried on the extensive and thorough research to the traffic road target recognition method based on remote sensing image. Based on remote sensing image edge detection, line based element extraction, low, medium and high resolution remote sensing image automatic extraction method, curve road automatic extraction method, based on remote sensing image of the bridge, airport runway and airport area automatic identification method.