Pose animation pdf

Look at what happens when a ball hits the ground. The force of the motion squashes the ball flat, but because an object needs to maintain its volume, it also widens on impact. This effect gives animation an elastic life-like quality because although it may not seem like it, squash and stretch is all around you. Squash and stretch imitates that and exaggerates it to create some fun. Check out the example below from the TV spot we did for Eastlink :.

When the letters spring from the ground, they elongate to show the impression of speed. Conversely, the letters squash horizontally when they come into contact with the ground. This conveys a sense of weight in each letter.

Do you swing your foot back to wind up? Steady yourself with your arms? Anticipation is the preparation for the main action. The player striking the soccer ball would be the main action, and the follow-through of the leg is well… the follow through. Notice how the progression of action operates in this scene. When filming a scene, where do you put the camera?

Where do the actors go? What do you have them do? The combination of all these choices is what we call staging. Staging is one of the most overlooked principles. In the clip below from our video for Sevanta Dealflow, the placement of the character within the scene allows us to quickly follow his actions and gives us a good sense of the physical layout. This motivates the cut to a medium shot, which then pulls back to the two-shot to reveal that his colleague is also worried about this problem.

It builds from problem to realization to shared understanding, to the beginning of a solution, all in a visual telling. These are two ways of drawing animation.

Straight ahead action is where you draw each frame of an action one after another as you go along. With pose-to-pose, you draw the extremes — that is, the beginning and end drawings of action — then you go on to the middle frame, and start to fill in the frames in-between.

Pose-to-pose gives you more control over the action. Poser is a 3D computer graphics program optimized for 3D modeling of human figures. It is known for enabling digital artists to easily produce 3D animations and digital images, as well as the extensive availability of third-party digital content.

Poser is a digital stage that gives you full creative control. Work with 3D figures for any project requiring rendered images or animated video. Poser is an ecosystem full of ready-to-pose 3D human figures, hair, clothing, props, scenery, lighting and cameras you need to bring your stories, dreams and fantasies of all genres to life. From historic to contemporary, sci-fi to fantasy, Poser is the 3D graphics software tool used by studios and hobbyists alike.

Follow Through and Overlapping Action Follow through and overlapping action is a general heading for two closely related techniques which help to render movement more realistically, and help to give the impression that characters follow the laws of physics, including the principle of inertia.

A third related technique is "drag", where a character starts to move and parts of him take a few frames to catch up. These parts can be inanimate objects like clothing or the antenna on a car, or parts of the body, such as arms or hair.

On them human body, the torso is the core, with arms, legs, head and hair appendices that normally follow the torso's movement. Body parts with much tissue, such as large stomachs and breasts, or the loose skin on a dog, are more prone to independent movement than bonier body parts.

Again, exaggerated use of the technique can produce a comical effect, while more realistic animation must time the actions exactly, to produce a convincing result. The "moving hold" animates between similar key frames, even characters sitting still can display some sort of movement, such as the torso moving in and out with breathing. Slow In and Slow Out The movement of the human body, and most other objects, needs time to accelerate and slow down.

For this reason, animation looks more realistic if it has more drawings near the beginning and end of an action, emphasizing the extreme poses, and fewer in the middle. Ball Fast to slow Slow out Pendulum Slow In Arc Most natural action tends to follow an arched trajectory, and animation should adhere to this principle by following implied "arcs" for greater realism. This can apply to a limb moving by rotating a joint, or a thrown object moving along a parabolic trajectory.

The exception is mechanical movement, which typically moves in straight lines. As an object's speed or momentum increases, arcs tend to flatten out in moving ahead and broaden in turns. An object in motion that moves out of its natural arc for no apparent reason will appear erratic rather than fluid.

Traditional animators tend to draw the arc in lightly on the paper for reference, to be erased later. Secondary Action Adding secondary actions to the main action gives a scene more life, and can help to support the main action.

A person walking can simultaneously swing his arms or keep them in his pockets, he can speak or whistle, or he can express emotions through facial expressions. The important thing about secondary actions is that they emphasize, rather than take attention away from, the main action. They allow extending the range of mo- enable precise control. The relation of skin strain to fine- tions beyond muscle-driven expressions by incorporating ex- scale wrinkles can be learned more efficiently than the re- ternal forces.

Essa et al. Sifakis et al. Fur- accuracy can be obtained by adding more poses. Exploiting this, we derive a highly parallel methods for wrinkle synthesis as a function of the strain GPU implementation that achieves real-time animations of measured in an underlying tissue model.

Those approaches detailed facial expressions. Those animations are controlled allow for realistic model behavior, but in turn require com- through a set of marker positions, which in combination with plex parameter tuning and expensive computations. Pighin et al. As such, our method provides a simple way to re-use cap markers, thus requiring an expensive nonlinear model captured or manually modeled high-res face data.

Furthermore, they require spatial and temporal dense capturing of geometric wrinkle constraints. In contrast, our hybrid framework ex- 2. Related Work ploits the quasi-static nature of fine-scale wrinkles by em- Modeling, acquisition, and animation of human faces are ploying an example-based model for fine-scale skin correc- topics largely explored in computer vision and computer tions, and thereby avoids spatio-temporal dense acquisition graphics [NN99,PL06].

Our target problem falls into the cat- and any expensive nonlinear optimizations at run-time. They typically combine a fast, linear skeletal sub- One large family of methods in face animation employs space deformation SSD [MTLT88] with a nonlinear pose- model blending. Blend shapes, dating back to the early work space deformation PSD [LCF00] that interpolates correc- of Parke [Par74] and commonly used in the industry today, tion vectors among example poses.

The recently presented method of [MA07] is related to our work in the sense that it computes high-resolution defor- Large-Scale Fine-Scale mations from a few handle vertices, but it focuses on the se- Linear Deformation Pose Space Deformation lection of handles given a large training dataset.

Other recent methods [WPP07, WSLG07] learn example-based correc- tions on sparse points and assume that these corrections can be smoothly interpolated. In general, any pose-space method requires the definition of a suitable pose space, which for SSD can easily be defined using bone transformations. In the context of face models, we define a novel pose space based on local skin strain and derive corresponding PSD and WPSD techniques for fine-scale skin correction.

In contrast, our pose-space representation allows to transfer only the fine-scale details from one character onto another, which can be used to add realistic wrinkles to existing models and face animations. Figure 2: Our hybrid face animation pipeline computes the 3. Overview large-scale facial motion from a linear deformation model, and adds fine-scale details using a pose-space deformation We start with an overview of the input data and the workflow that learns skin corrections from a set of example poses.

To this end, of the animation by deforming F from its rest pose. Its deformation is out the need to produce a consistent mesh e.

The resulting large-scale deformation successfully according to the deformed feature graph, and use this infor- captures the overall facial expression, but lacks fine-scale fa- mation for a pose-space correction of fine-scale facial details cial details such as expression wrinkles Fig. The nonlinear behavior that goes beyond the linear large- scale motion is learned in a preprocess from a set of P 4.

Large-Scale Deformation example poses. Given the example poses, the scale deformation u is computed by minimizing a simplified, corresponding fine-scale details are extracted as the differ- quadratic thin shell energy. This amounts to solving the cor- ence between the examples and the results of the large-scale responding Euler-Lagrange equations deformation for the same poses, and are stored per-vertex in local tangent frames.

Figure 3: Wrinkle Formation and Skin Strain. At run-time, the a measure of strain. How- ample pose to the input pose. Alterna- duced by expression wrinkles, cannot be reproduced. These tively, a per-vertex strain tensor could be used, but in practice fine-scale deformations are highly nonlinear with respect our discrete strain approximation turned out to be sufficient.

However, they vary smoothly as a function of facial pose, hence we have opted 5. Pose-Space Deformation for learning, as a preprocess, the fine-scale displacement d from a set of example poses, and then at run-time compute In order to exploit the connection of skin strain to wrin- it by interpolation in a suitable facial pose space. To this end, we We first define a facial pose space based on a rotation- represent each facial expression by its rotation-invariant fea- invariant feature vector of skin strain Section 5.

Hence, each mulate the learning problem as scattered-data interpolation facial expression corresponds to a point in an F-dimensional in this pose-space Section 5.

We extend the basic method pose space, which constitutes the domain of the function we to weighted pose-space deformation Section 5. Its range is the fine-scale detail correction d, allowing for a more compact basis.