Table 4.

Advantages and disadvantages of presented techniques.

	Advantages	Disadvantages
Hand-crafted motion features	- There is no need of a large amount of data for training. - It is simple and unambiguous to understand the model and analyze and visualize the functions. - The features used to train the model are explicitly known.	- Usually these features are not robust. - They can be computationally intensive due to the high dimensions. - The discriminative power is usually low.
Depth information	- The 3D structure information of the image that depth sensors provide is used to recover postures and recognize the activity. - The skeletons extracted from depth maps are precise. - Depth sensors can work in darkness.	- Depth maps have no texture, making it difficult to apply local differential operators. - The global features can be unsettled because depth maps may contain occlusions.
Deep Learning	- There is no need of expert knowledge to get suitable features, reducing the effort of feature extraction. - Instead of designing them manually, features are automatically learned through the network. - Deep neural networks can extract high-level representation in deep layer, making it more suitable for complex tasks.	- Need to collect massive data, consequently there is a lack of data sets. - Time consuming. - Problem of models capability of generalization.