Systematic Evaluation of Spatio-Temporal Features on Comparative Video Challenges

摘要: In the last decade, we observed a great interest in evaluation of local visual features domain images. The aim is to provide researchers guidance when selecting best approaches for new applications and data-sets. Most state-of-the-art have been extended temporal allow video retrieval categorization using similar techniques those used However, there no comprehensive these. We first comparative based on isolated well defined alterations data. select three most promising approaches, namely Harris3D, Hessian3D, Gabor detectors HOG/HOF, SURF3D, HOG3D descriptors. For detectors, measure their repeatability challenges treating videos as 3D volumes. To evaluate robustness spatio-temporal descriptors, propose principled classification pipeline where increasingly altered build set queries. This allows an in-depth analysis descriptors combinations.