Table 1.

Comparison of different annotation tools with their different focuses.

Approach	Idea	Method	Pros and Cons
Our approach	Suggest labels based on a small subset of annotations. In depth analysis of different preprocessing methods and variants of dynamic time warping.	The main focus is the analysis of different variants of dynamic time warping used for label-, and clustering-suggestions. The methods were applied to wrist worn sensors mostly. Evaluation was done in regards to time offset to the correct label and recall of the method.	An in-depth analysis among different datasets with different configurations. On the downside the tool itself is just a prototype, usability of the tool is not tested.
Label Movie [18]	Designing a complete multimedia annotation tool with automatic annotation and crowd-sourcing capabilities.	Used dynamic time warping and SVM time series prediction with a focus on usability of the application. Classification results shown in a Gram matrix to the user. Focus on crowd-sourcing capabilities with the combination of domain experts’ and technical experts’ knowledge.	The tool is fully developed with a lot of functionality, especially the capability for crowd-sourcing. On the downside the evaluation of the tool is lacking in detail and it is not publicly available.
Multimodal Multisensor Activity Annotation Tool [19]	A multimodal annotation tool that is able to handle multiple sensor types like video, depth, and body worn sensors.	The focus is put on capturing many different types of sensors and displaying them in a useful fashion. In contrast to the other methods, this tool is able to capture sensors live and synchronize them. Capabilities for automated annotation are present, but not implemented yet.	Live capturing of different types of sensors is integrated and the tool seems to be designed very concisely. However, as of yet automatic annotations are not integrated though the architecture allows for that.
Smart Video Browsing [12]	Using clustering methods, automatically segment videos into different parts to improve navigation within a video.	For clustering the tool uses color and motion features to distinguish different parts of the video. These can be browsed by the user to distinguish different parts of the video.	The tool does not rely on pretrained methods and can thus easily be used. It does not, however, provide automatic labeling functionality.