Video stitching system of heterogeneous car video recorders

Abstract

Many heterogeneous car video recorders are sold in the market. In order to record the ultra-wide-angle road scenes, at least two heterogeneous recorders are used. These heterogeneous recorders have different viewing angles, different resolutions, and different lens sensors. Because of different hardware and software of the heterogeneous recorders, the captured videos are heterogeneous, which is very challenging for video stitching research. The traditional image stitching system includes color correction, feature detection, feature descriptor, feature matching, and video stitching. When a traditional image stitching system is used to process the images captured by homogeneous cameras, the results are reasonably good. However, when a traditional image stitching system is used to process the images captured by heterogeneous cameras, the results are not so good. Furthermore, applying a traditional image stitching system to process videos captured by heterogeneous car video recorders is timeconsuming. This paper presents a study that tested multiple existing methods to evaluate their capability when used to stitch heterogeneous images to allow a driver to see an ultra-wide-angle driving view without blind spots and to record these images. Experimental results show that some methods used in this study can be used for this purpose, but they have significant error rates and are timeconsuming.