New Approach of Generating Ground-Truth for Local Surveillance Dataset Tested with Benchmark Background Subtraction Models

Abstract

Background subtraction is the dominant approach in the domain of moving object detection. Lots of research has been done to design or improve background subtraction models. However, there are a few well-known and state-of-the-art models that can be applied as a benchmark. Generally, these models are applied to different dataset benchmarks. Most of the time, choosing an appropriate dataset is challenging due to the lack of dataset availability and the tedious process of creating ground-truth frames for the sake of quantitative evaluation. Therefore, in this article, we collected local video scenes of a street and river taken by a stationary camera, focusing on dynamic background challenges. We presented a new technique for creating ground-truth frames using modeling, composing, tracking, and rendering each frame. Eventually, we applied three promising algorithms used in this domain: GMM, KNN, and ViBe, to our local dataset. Results obtained by quantitative evaluations revealed the effectiveness of our new technique for generating the ground-truth scenes to be benchmarked with the original scenes using a number of statistical metrics.