RGB-D Tracking Based on Kernelized Correlation Filter with Deep Features
This paper proposes a new RGB-D tracker which is upon Kernelized Correlation Filter(KCF) with deep features. KCF is a high-speed target tracker. However, the HOG feature used in KCF shows some weaknesses, such as not robust to noise. Therefore, we consider using RGB-D deep features in KCF, which refer to deep features of RGB and depth images and the deep features contain abundant and discriminated information for tracking. The mixture of deep features highly improves the performance of the tracker. Besides, KCF is sensitive to scale variations while depth images benefit for handling this problem. According to the principle of similar triangle, the ratio of scale variation can be observed simply. Tested over Princeton RGB-D Tracking Benchmark, Our RGB-D tracker achieves the highest accuracy when no occlusion happens. Meanwhile, we keep the high-speed tracking even if deep features are calculated during tracking and the average speed is 10 FPS.
KeywordsRGB-D KCF Deep features Scale estimation
This work is supported by the National Natural Science Foundation of China (No. 61273273) and by Research Fund for the Doctoral Program of Higher Education of China (No. 20121101110034).
- 1.Awwad, S., Piccardi, M.: Prototype-based budget maintenance for tracking in depth videos. Multimedia Tools Appl. 1–16 (2016)Google Scholar
- 2.Bibi, A., Zhang, T., Ghanem, B.: 3D part-based sparse tracker with automatic synchronization and registration. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1439–1448 (2016)Google Scholar
- 4.Danelljan, M., Shahbaz Khan, F., Felsberg, M., Van de Weijer, J.: Adaptive color attributes for real-time visual tracking. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1090–1097 (2014)Google Scholar
- 5.Hannuna, S., Camplani, M., Hall, J., Mirmehdi, M., Damen, D., Burghardt, T., Paiement, A., Tao, L.: DS-KCF: a real-time tracker for RGB-D data. J. Real-Time Image Proc. 1–20 (2016)Google Scholar
- 7.Kang, K., Li, H., Yan, J., Zeng, X., Yang, B., Xiao, T., Zhang, C., Wang, Z., Wang, R., Wang, X., et al.: T-cnn: Tubelets with convolutional neural networks for object detection from videos. arXiv preprint arXiv:1604.02532 (2016)
- 10.Nam, H., Han, B.: Learning multi-domain convolutional neural networks for visual tracking. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4293–4302 (2016)Google Scholar
- 11.Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014)
- 12.Song, S., Xiao, J.: Tracking revisited using rgbd camera: Baseline and benchmark. arXiv preprint arXiv:1212.2823 (2012)