Automatic offside determination in soccer videos based on improved GMM-FFD-FDSST

To solve the problems of target holes, false detections, edge missing, occlusion, and lighting changes in offside detection in football videos, this study proposes an improved Gaussian Mixture (GMM) model combined with Four Frame Difference (FFD) for automatic offside detection. This method introduces an adaptive learning rate mechanism to dynamically adjust the update rate of the background model, and improves the continuity and integrity of the moving target edges by integrating the Canny edge detection operator with the four frame differential results. In the target tracking stage, this study further combines Kalman filtering with Fast Discriminant Scale Space Tracking (FDSST) and adopts a dynamic switching mechanism to address the issues of target occlusion and loss. The experimental results show that the practical advantages of the proposed method in offside determination of football videos are mainly reflected in the following three aspects. Firstly, by introducing adaptive learning rate and Canny edge fusion strategy, the average detection accuracy was improved to 92.9% under conditions of sudden lighting changes and background interference. Secondly, the combination of FFD and dynamic thresholding strategy effectively suppresses background noise, with an average false detection rate as low as 5.1%, significantly lower than similar comparison methods. Thirdly, the dynamic switching mechanism between Kalman filtering and FDSST algorithm can maintain stable tracking performance even when the target is severely occluded, and the center point error is always controlled within a small range. The above improvements enable the proposed method to have high accuracy, low false alarm rate, and strong robustness in practical competition scenarios, and have good application and promotion value.