Video datasets and multiple angles

[Michael O'Brien]

When creating a datasets comprising of variable length video sequences of sports action does having multiple camera angles help or hinder training classifiers?

to elaborate the dataset does not necessarily showing the same exact sequence from multiple angles but each example would be recognised by an informed human operator as showing an example of class label 

Would it be recommended that any dataset include as little variation in position and angle of the camera recording the action or does the "rawness" of the variation help classification?

Michael