Soccer games’ automatic offside determination technology based on video analysis

Abstract

To better define soccer offside, a Gaussian mixture model and an automatic four-frame difference offside detection method are both adopted to avoid problems such as target holes, false judgment, and missing edges in final results. In addition, this study also upgrades the method utilizing Kalman filtering with fast discriminative scale space tracking algorithm to realize continuous and stable tracking of moving targets. According to results, the time used for single-frame detection with the four-frame difference method in three scenarios is 245 ms, 317 ms, and 278 ms, respectively. The accuracy of the improved method is from 85.4% to 99.3% and the sequence number of frames is from 5 to 225. Among the average accuracies of the four methods in the three scenarios, the lowest lies in the three-frame difference method at 50.3%, while the four-frame difference algorithm has the highest value of 92.9%; in terms of the average false detection rate of the four methods, the highest is 41.1% for the Gaussian hybrid algorithm, and 9.4% for the three-frame difference method. In addition, the improved algorithm has the lowest average mis-detection rate of 5.1%. The average accuracy of the improved algorithm is 90.8% for scene 1, 92.9% for scene 2 and 95.2% for scene 3. In comparison, the traditional fast discriminative scale-space tracking algorithm runs at 40 frames per second, but the improved algorithm runs at 35 frames per second. Therefore, the improved motion target detection algorithm and tracking algorithm are more robust, which is significant to sports development.