There are two versions of optical flow: Feature based (sparse) or dense. In the dense version this is done for all the image pixels, while in the sparse one only certain characteristic feature points are tracked. However, both approaches depend on the tracking of pixel quantities. This is fundamentally different than tracking the whole patch, because in return one obtains a full set of pixel-level correspondences. This is not only a vague statement, but rather carries the idea that the correspondence estimation is not constrained. Remember in the motion case, one would assume that the motion is "rigid" or "articulated". And imposes this prior into the tracking framework. If optical flow is used to estimate the dense trajectories, than this constraint is not assumed and one could as well track the deformable bodies, regardless of the deformation model.

Optical flow provides you more freedom and information about the tracked scene. Yet, my experience is that the motion tracking methods are more robust and reliable. But keep in mind that, by applying human models or temporal models on top of such tracked trajectories (from OF) researchers are capable of developing robust articulated / deformable / model-based tracking algorithms.

There are two versions of optical flow(OF): Feature based (sparse) or dense. In the dense version OF is applied to all the image pixels, while in the sparse one, only certain characteristic feature points are tracked. However, both approaches depend on the tracking of pixel quantities. This is fundamentally different than tracking the whole patch, because in return one obtains a full set of pixel-level correspondences. This is not just a vague statement, but rather carries the idea that the correspondence estimation is not constrained. Remember in the motion case, one would assume that the motion is "rigid" or "articulated" and imposes this prior into the tracking framework. If optical flow is used to estimate the dense trajectories, than this constraint is not assumed and one could as well track the deformable bodies, regardless of the deformation model.

Optical flow provides you more freedom and information about the tracked scene. Yet, my experience is that the motion tracking methods are more robust and reliable. But keep in mind that, by applying human models or temporal models on top of such tracked trajectories (from OF) researchers are capable of developing robust articulated / deformable / model-based tracking algorithms.

There are two versions of optical flow: Feature based (sparse) or dense. In the dense version this is done for all the image pixels, while in the sparse one only certain characteristic feature points are tracked. However, both approaches depend on the tracking of pixel quantities. This is fundamentally different than tracking the whole patch, because in return one obtains a full set of pixel-level correspondences. This is not only a wage statement, but rather carries the idea that the correspondence estimation is not constrained. Remember in the motion case, one would assume that the motion is "rigid" or "articulated" and etc. And imposes this prior into the tracking framework. If optical flow is used to estimate the dense trajectories, than this constraint is not assumed and one could as well track the deformable bodies, regardless of the deformation model.

Optical flow provides you more freedom and information about the tracked scene. Yet, my experience is that the motion tracking methods are more robust and reliable. But keep in mind that, by applying human models or temporal models on top of such tracked trajectories (from OF) researchers are capable of developing robust articulated / deformable / model-based tracking algorithms.

There are two versions of optical flow: Feature based (sparse) or dense. In the dense version this is done for all the image pixels, while in the sparse one only certain characteristic feature points are tracked. However, both approaches depend on the tracking of pixel quantities. This is fundamentally different than tracking the whole patch, because in return one obtains a full set of pixel-level correspondences. This is not only a vague statement, but rather carries the idea that the correspondence estimation is not constrained. Remember in the motion case, one would assume that the motion is "rigid" or "articulated". And imposes this prior into the tracking framework. If optical flow is used to estimate the dense trajectories, than this constraint is not assumed and one could as well track the deformable bodies, regardless of the deformation model.

Optical flow provides you more freedom and information about the tracked scene. Yet, my experience is that the motion tracking methods are more robust and reliable. But keep in mind that, by applying human models or temporal models on top of such tracked trajectories (from OF) researchers are capable of developing robust articulated / deformable / model-based tracking algorithms.