Their "theory" is lovely in Sections 1.1.1 and 1.1.2, which, however, lack any references to any serious literature [1]. Then, Table 2 confirms almost universal and severe limitations. Yet, they conclude that "emotion recognition is crucial in multiple fields, including healthcare, E-learning, online shopping, etc.".
[1] The paper of Ekman is serious, but someone should dig through the 12881 citations it has gathered; much of these are criticism.