A post-processing scheme for the performance improvement of vehicle detection in wide-area aerial imagery

Abstract

In this paper, we present a post-processing scheme to improve the performance of vehicle detection in wide-area aerial imagery. Using low-resolution aerial frames for the performance analysis, we adapted nine algorithms for vehicle detection. We derived a three-stage scheme to measure performance improvement on the selected five object segmentation algorithms before and after post-processing. We compared automatic detections results to ground-truth objects, and classified each type of detections in terms of true positive, false negative and false positive. Several evaluation metrics are adopted for the experimental study.