Abstract— This paper presents an LSF-based framework for detecting arcs in broadcast soccer video. The successful identification of the arcs will evidently facilitate the soccer video analysis. The existing methods are not available for all playfield arcs including both the middle field circle and penalty box arcs. A new algorithm is proposed in the paper improved from LSF, called ALSF (Advanced Least Square Fitting), which can be used to detect arcs even though they are only 1/4 of eclipses (the same as penalty box arcs). With the improvement, we proposed a new framework to detect the arcs in broadcast soccer video. The proposed method first removes the points of straight lines. Then, all points of each connective area are transformed into a new reference frame to use LSF to get the equations of arcs. Experiments on more than 3 hours broadcast soccer video show the proposed method is effective with above 98% precision and 70% recall. I. INTRODUCTION With the development of the interactive video review, automatic soccer video analysis give fans a fast and convenient way to access the match rather than watch the game themselves. Much effort done in the field, the subject remains an open issue. The reason is that the structure information is scarce and the usable information (the lines, the arcs, the persons and the ball, etc) is difficult to be extracted preciously and quickly, which harms the accuracy and practicality of the application. Playfield arcs detection can significantly facilitate the soccer video mining, such as playfield registration as well as the camera calibration [1-3]. As we know, the arcs are unique in one playfield. For example, the left orientation arc with ends on a line must be the left penalty box arc, and the whole eclipse or arc with ends on the edge of screen must be the middle field circle. So if we can get an arc in one frame, the position will be known. With the detection results, more feature points (the points on both a straight line and an arc) can be gotten for calibration. More positions, such as the positions near the penalty box, can be restored. The dominative algorithms of arc (eclipse) detection can be divided into three categories: Eclipse Hough Transformation and its variations [4-14], Least Squared This work is supported by the National Natural Science Foundation of China under Grant No. 60573167 and Intel Cooperation Project: Personal Desktop Video Mining System. Fei Wang is with Department of Computer Science and Technology, Tsinghua University (fei-wang04@mails.tsinghua.edu.cn ). Bo Yang is with Department of Computer Science and Technology, Tsinghua University. Lifeng Sun is with faculty of Department of Computer Science and Technology, Tsinghua University. Shiqiang Ynag is with faculty of Department of Computer Science and Technology, Tsinghua University Fitting [15], and Invariant Pattern Filter [16]. Nowadays, the EHT is widely used to detect the eclipses in soccer video. The SEHT [6] (standard eclipse Hough transformation) uses each edge points to vote for all possible eclipse equations, which needs large computation and storage consumption (because of 5 variants and the complexity is O(n 5 )). The Randomized Eclipse Hough Transformation [7] improves SEHT, but the aberration of the arcs in soccer video will make it fail. To overcome the shortcomings of the SEHT, some research work has been proposed for detecting eclipses in global views of soccer videos [3-5]. These algorithms can only detect almost whole eclipses or eclipses more than their half in the frames with the middle lines, which is very restricted and cannot take effect on the arcs of penalty boxes. LSF [15] and Invariant Pattern Filter [16] are faster and more widely available than above algorithms, which may be able to detect the penalty box arcs, but they also fail in the soccer video analysis. The key obstacle why they are not available for soccer video is the arcs will have errors while shooting. The two algorithms need a lot of points to eliminate them. But in broadcast soccer video, we usually can not get enough right points, because the line point extraction is not able to get all the points of lines in one frame. More seriously, the person or other objects may occult some parts of the lines, and then the arcs will become to short line-lets whose points are not enough for LSF. Another possible situation is the arcs and other objects are connective, so we can not tell points of the arcs from points are of the objects. The LSF is more robust than Invariant Pattern Filter when we cannot get enough points, but there is one problem of LSF if we use it in the soccer video. In the LSF algorithm, we must solve a set of linear equations for get the equation of arcs. When there are a lot of points in one part, the linear equations will become ill-conditioned, because some parameters of the parameter matrix will be much bigger than others. So the correct results cannot be achieved in the broadcast soccer video. To overcome the shortcomings of LSF, two problems must be solved. Firstly, we must separate the points of arcs from others. This process will decrease error introduced by points which are not contained by the target arc. The difficult is that other objects removed, the arcs will be affected, even though they will be segmented into several parts which have no enough points. Secondly, we must use some measures to decrease the gap among the parameters of the parameters matrix of linear equations in the process. Then the correct results can be achieved. In this paper, we propose a new framework to detect the Fast Arc Detection Algorithm for Play Field Registration in Soccer Video Mining Fei Wang, Lifeng Sun, BoYang, Shiqiang Yang