| Radiotherapy, surgery and chemotherapy are three most important cancer treatment means, and there are estimated to be more than60%～70%cancer patients who need radiation therapy, including radiotherapy alone, preoperative or postoperative radiotherapy and radiotherapy combination with chemotherapy. The purpose of the radiotherapy is maximizing tumor local dose, improve tumor local control, and minimize the damage for the normal tissues, so as to improve the cure of cancer and the patients has a quality of life. The development course of radiotherapy as follow conventional radiotherapy,3DCRT(three-dimensional conformal radiotherapy), IMRT(Intensity Modulation radiotherapy) and IGRT(Image Guided radiotherapy accurate radiotherapy).Fractional irradiation strategy was take during radiotherapy for protects the healthy tissue. Even if a good position fixed technology, can guarantee the treatment of repeatability of body posture, but the patient’s breathing and organs exercise that cause target area and endanger organs movement, which make them deviation from the radiation fields, cause tumor dose inadequate, normal organs doses excessive. For image guided radiotherapy for age, this has taken more and more attention. Related test shows that the movement produces serious effect. If we can control the influence of breathing in radiotherapy, which can reduce the expansion in edge, narrow treatment error, improve the accuracy.At present, research and application of control the breath movement has many methods, such as active breathing control technology, breathing guide door control technology, forced breathless technology and synchronization breath control technology (CyberKnife), etc. In ideal conditions, if we can continuous real-time tracking, its treatment precision can reach1mm. This way demanding:real-time identification in the location of the tumor in the body; Consider the response of the system time delay, and expected tumor location change degree; Constantly correction and adjust the output of ray; In order to adapt to the volume of the lung respiratory and structure of critical changes, adjust the output dose. This is the most promising way for radiation therapy, which can completely solve the impact on the breathing movement a, so it was be widely study.The purpose of this paper is introducing the computer stereo vision technology into real-time tracking of breath movement. Use of binocular vision technology acquisition the markers image which on the patient’s abdomen surface, and then real-time measuring the marker’s three dimensional space coordinates, send the information to reverse tracking platform system, so as to achieve the breathing movement reverse compensation, and improve the radiation treatment effect.In order to achieve the purpose of real-time tracking, the concrete methods for overview:The respiratory movement is tracking by two cameras composed of computer vision system, which is real-time matching a marker’s coordinates in the two image acquisition by the cameras, who is place in the abdomen surface, and then according to the basic principle of binocular calculated the three-dimensional coordinates of the marker. Combining time parameters calculated the feature points3d coordinate changes, respiratory movement is tracking by the real-time. SITF (Scale Invariant Feature Transform) was used as the registration algorithm, which has strong robustness and accurate matching performance, at the tracking process. In order to reduce time overhead, dynamic choice matching image and local searching strategy are used in the process of calculation and mergence. The system including the following aspects:camera calibration, marker feature extraction, marker dynamic tracking, binocular vision to match and the markers’3D information calculations, etc.Camera calibration is the foundation and the key point of computer vision, it establish a basic contact between the object and the images. Stereo vision imaging principle is an imitation of the human eye, through the lens Samples of the object for2d image, which according to the same object in different camera images, and then calculate the object3d coordinate information. The camera external parameters are representative of camera position and orientation relative to an actual coordinate transformation of coordinates, internal parameters of the optical nature are representative of camera. The calibration is based on the previously coordinates of3d space point, through reverse calculation to determine the camera system’s internal and external parameters. In this system the camera calibration scheme is separately to two cameras are single calibration, and then to stereo calibration. The single camera calibration is derived the internal parameter matrices, distortion coefficient, the rotation matrix R and peace move vector T. The first two constitute the camera intrinsic parameters; R and T constitute the object position and direction of the camera parameters. In three-dimensional calibration, the rotation matrix R and vector T are use to contact the two camera. In the broadest sense, the camera calibration can be divided into three types:traditional calibration approach, based on active visual calibration method and the self calibration.2D (plane of the board) traditional calibration approach was used in this experimental system. The detail of the calibration theory, calibration test and the results has been described in the paper.The tracking effect of computer vision depends on the reliability of the video image and image registration speed. One of the key point is the computer can be quickly identified in the target in the images. The key to achieve the target tracking according to complete segmentation, reasonable target characteristics extracted and accurately identify targets, also the algorithm time must to be considered. Based on motion analysis method and based on image matching are the two method of target tracking. Light flow separation method and frame difference method are the main method of kinematic analysis. Based on the image matching method is use the basic idea of the image registration, matching the same as object in the different picture, and analysis the position relative and movement of the target. The correct intercept probability and positioning accuracy of image matching are the main performance indexes. The picture registration can be divided into the area matching, model matching, frequency domain and matching feature matching. Some scholars test shows that:in view of the different scene, the illumination change, image geometric distortion, resolution differences, rotation, fuzzy and six species of the image compression, SIFT descriptor performance is the best. Therefore, the system using SIFT as image matching algorithm in the target tracking.SIFT algorithm produces the image characteristics has the following main steps:(1) extreme value scale space detection;(2) the key points of the positioning;(3) to determine the direction;(4) describe key points. Space theory is simulated image data multi-scale features, the main process is extreme value detection:in the space use the only one linear nuclear--gaussian kernel, establish a gaussian pyramid. Therefore, the theory of scale space main idea is using the gaussian check the original image scale change, get the multi-scale images of scale space said sequence, for these series feature extraction scale space. Each key point has the information: position, the scale, the direction. After the characteristic of the two image characteristic vector generation SIFT, characteristic vector Euclidean distance of the key point as two images of similarity measure judgment. Take a key point in the image1, and find out the nearest key point by the Euclidean distance in the other image. This paper use C++realized the algorithm, and the relevant match test.Binocular computer vision tracking system is based on the human eye binocular parallax measuring principle, through the two cameras to different angles on the same object with parallax for the two images, and then based on two of the2d image reduction of the3d object information. Determine the position of the relationship between two cameras is very important about measuring accuracy, and this position relationship can be obtained through three-dimensional calibration. This paper introduces the calibration method and relevant test results.System software engineering subdivided into the following modules:system initial module, video image calibration module, markers image SIFT feature extraction module, stereo matching module,3d coordinate calculation module and track realize module. Initial module:the main function is stereo camera calibration, separately for obtain parameter matrices, distortion coefficient, the rotation matrix R and peace move vector T of the camera. Video image calibration module:with calibrated parameters, each frame of the video image will be calibrating the distortion. Markers image SIFT feature extraction module and stereo matching module:get matching point of the markers in each video image.3d coordinate calculation module:combined with (xl,yl) and (xr,yr), which has been obtained, calculates the markers3d coordinate.3d movement calculation module:introducing time function, calculate the markers real-time speed. The subsequent frames registration image choice:in order to reduce computation cost, and improve the processing speed, some image registration processing was been adopted. Among them SIFT algorithm, after the first time complete markers image sampling and match, the subsequent video image use the markers position as the center of the first frame, dynamic selection the area for the next match registration, and then get accurate position is mapped to the original image. The selection of dynamic region size, decided the video matching efficiency.In chapter5, the author introduces the hardware and software of binocular vision tracking platform, and completed the related test. Experimental results show that in a breathing cycle, the influence of the biggest sports scope is not much more than0.2cm between the vision measuring and practical measurement, and can achieve real-time calculation. The errors in calculation mainly come from the following aspects:the camera parameters and the actual internal error; camera pixel restrictions. Conclusion:The proposed method is characterized by good robustness and effectively achieves real-time multi-target tracking in complex occlusion scenes for respiratory movement tracking. In the final, we also discuss some problems which haven’t been completed; in addition, we prospect some work which we’ll do in the future. |
PDF Full Text Request