Research on Ball Strike Judgement Improvement Using AKAZE Algorithm under Multi-perspective Video Fusion

Addressing the issue of judgment bias in baseball games due to view occlusion from a single camera angle, this study proposes the use of multi-camera video fusion and real-time motion monitoring technologies to merge the visual information from two cameras into a unified video frame, thereby enhancing the judgment accuracy. To achieve faster and more precise image matching, an improved AKAZE algorithm for efficient image registration is proposed. Firstly, in the feature extraction phase, the AKAZE algorithm constructs a nonlinear scale space and uses adaptive descriptors to effectively describe feature points. Techniques such as superpixel segmentation and illumination adjustment are employed to better adapt to complex environmental conditions. Secondly, the FLANN method is used for pre-matching the feature points, quickly calculating the nearest neighbors for each feature point to accelerate the matching process. The ratio test proposed by Lowe is used to compare the distance differences between the nearest and next-nearest neighbors, discarding unlikely matching feature points. Finally, further selection of matches is performed using the RANSAC algorithm to remove erroneous matches, thus improving the accuracy and reliability of the final match. To validate the effectiveness of the proposed method, experiments were conducted using a real-time video stream dataset collected by cameras, compared with the SIFT, ORB, and AKAZE algorithms. The results showed that the proposed method maintains high accuracy while achieving rapid image registration and demonstrating robustness under changes in illumination, blur, and compression transformations. The image fusion algorithm enables seamless visualization and tracking of the transferred attention category across different camera views, meeting the real-time detection needs for ball strike judgements in baseball.