Limitations with measuring performance of techniques for abnormality localization in surveillance video and how to overcome them?