Superpixel-based statistical anomaly detection for sense and avoid

This paper presents a novel preprocessing method for detecting small objects of interest within a high-resolution image, applied to the problem of visually detecting possible aircraft collisions (Sense and Avoid) for UAV platforms. The method is based on superpixel image segmentation combined with subsequent statistical analysis and anomaly detection. The existence of a possible target within a superpixel is described in terms of how it affects the local superpixel statistics and this signature statistical profile is consequently used to identify regions of interest throughout the image. The approach eliminates upwards of 90% of the total image area, significantly reducing the workload of further processing stages.