Acoustic cameras are increasingly used for continuous, non-intrusive recording and counting of fish passage in natural environments and artificial structures such as fishways. However, analysing the large number of videos recorded is time consuming. Although automatic reading processes have been developed, the poor quality of acoustic images, including discontinuity of the signal for a single object, is challenging. We developed an innovative method for analysing acoustic videos. Unlike previous methods, it focuses on swimming locomotion instead of the morphological properties of fish. Each image of a fish is pre-processed to remove discontinuities and restore the entire fish body as a single cluster of pixels. The set of pixels is then tracked to retrieve movement, independent of the displacement of the fish, using a mesh and a solid deformable model. The deformation to which the mesh is subjected between each pair of frames (i.e., deformation of the fish body) is summarised in a deformation map for each fish passage. Testing the method using a dataset of four species strongly suggested that deformation maps are species-dependent. These results must be extended to other species to confirm the effectiveness of the method for automatic identification of fish species and characterisation of their behaviour using acoustic camera records.