Standard rectilinear lenses of normal videos and images can only achieve a view angle of up to 90-100 degrees. Videos and images taken with fisheye lenses have a wide view angle of up to 180 degrees. The benefit of wide view angle of fisheye lenses is however comprised by significant barrel distortions, leading to visibly misshapen human faces and curve straight lines as objects move during a video sequence. The distinctive curvilinear perspective generated by fisheye lenses may be desirable in still photography; however in fisheye videos the resulting distortion renders them hard to understand, uncomfortable to watch and diminishes the overall continuity of the footage.
Engineers at Swansea University have developed an efficient and robust fisheye video correction pattern using per frame transformations which minimise time-delaying distortions while maintaining continuity of object movements in films. To obtain natural looking results, a number of distinct but related correction criteria are used to remove visual distortion such as straight line distortion, boundary consistency, orientation consistency, local shape similarity, homogeneity and chronological coherence. It also introduces automatically generating continuous and compatible correction constraints for individual video frames and presents a feasible implementation based on video tracking with a mechanism of adaptive error detection and correction. The input video is corrected by a method of streaming that works on a frame by frame basis.
The potential use of this new technique has been recognised but not restricted to mobile phones, surveillance monitoring, machine vision and biomedical scanning devices. With this system continuous video is produced comparable in visual quality to photographs produced by state of the art fish eye still photography cameras. This fish eye lens is likely to be popular amongst professionals within the film industry and armature film makers alike.
Inventor: Dr. Chenfeng Li