A Novel Multitouch Recognition Algorithm for Infrared Touch Screens Based on Geometric Construction and Clustering

Both sensor layout and sensor number determine the scan line density and limit the performance of modern infrared touchscreens in terms of recognition accuracy, positioning error, minimum touch object size, response time, and maximum number of touch points. This article explores methods to optimize the multitouch recognition algorithm, improving touchscreen performance without increasing hardware costs. To efficiently and accurately calculate touch point locations, a multitouch recognition algorithm based on geometric construction and clustering is proposed. The algorithm divides the set of intersection points into core, noncore, and noise regions by optimizing the median of blocked scan lines. A K-d tree is constructed for the intersection set using fast library for approximate nearest neighbor (FLANN), followed by the application of the KNN-BLOCK-DBSCAN clustering algorithm. Touch point coordinates are determined by merging and splitting clusters based on their density reachability. Experimental results show that our algorithm outperforms the polygon-cutting method in recognition accuracy, minimum touch object size, and response time. The response time is reduced by over 50%, with a minimum touch object diameter of 4 mm in the center and 5 mm in the corners, while eliminating touch-insensitive areas across the entire touch area.