Optical Flow

Prepared By
Aounaiza Ahmed
L20-2055
National University of Computer and Emerging
Sciences, FAST, Lahore.
Contents
Introduction:....................................................................................................................................2

Literature Review:...........................................................................................................................2

Methods and Results:.......................................................................................................................3

Discussion and Conclusion:.............................................................................................................6

Bibliograhy:.....................................................................................................................................7
Introduction:
Optical flow is an apparent movement pattern of objects, surfaces and rims in a visual
scene generated by the relative movement of an observer with a scene. Optical flow may
alternatively be described as the distribution of perceived brightness pattern motion velocities in
an image. In the 1940s the American psychologist James J. Gibson created the idea of optical
flow to describe the visual stimulation that animals move across the world. Gibson highlighted
the significance of optical flow for perception of affordability and the capacity to identify
environmental potential for action. Gibson's followers have further proven its ecological
approach in psychology the importance of the optical flow stimulus to the perception of motion
in the world by observers, the perception of the form, distance and movement of objects in the
world and the management of locomotive systems.

Literature Review:
As a significant part of optical flow research the motion assessments and video
compressions have been created. While the field of optic flow is superficially similar to a dense
field of motion derived from movements, optical flow is a study not only of how the optical flow
field itself is determined, but also of what it is used to estimate the three-dimensional nature and
structure of the scene, and how the 3D movement of objects and the viewers relate to a scenario.
The robotics investigators have utilised optical flow for a variety of fields such as the
identification and follow-up of objects, the dominating image extraction of a plane. Optical flow
information for controlling micro-air vehicles has been identified as beneficial. Consider a
cinematic clip of a ball travelling from the lower left side of a visual field to the top right side.
Motion estimation methods can establish that the ball moves up and down on a two-
dimensional plane and vectors may be derived from a succession of frames. The sequence is now
as stated for video compression reasons (e.g. MPEG). In the field of vision for machinery,
however, it remains unknown and still crucial whether the ball moves to the right or if the
observer moves to the left. We cannot say with confidence that the ball was travelling to the right
even if a static, patterned background were present in these five frames, since the pattern may be
infinitely far away from the observer. An optical flow sensor is a vision sensor that can monitor
optical flow or visual movement and produce optical flow measurements. Different optical flow
sensors designs are available. A picture sensor chip linked to a CPU that is configured to execute
a flow algorithm is a configuration. Another option employs a vision chip, a small
implementation of the integrated circuit with both the image sensor and the CPU on the same
die. A generic mouse optical sensor used in an optical mouse is an example of this. In some
instances, analogue or mixed signalling circuits can be used to provide rapid optical flow
calculations with minimum current usage.
Horn and Schunck proposed the constancy of brightness and the spa tial smoothness
limitations for estimating the optical flow, however their quadratic formula assumes Gaussian
statistics and is not resistant to outlying elements due to reflection, occlusion, limiting
movement, etc. Black and Anandan have proposed a strong estimating framework to cope with
such outliers, but they have not tried to represent the real brightness constancy and flow
derivatives statistics. Fermu did not seek to draw from instances flow statistics, however, by
analysing the effects of noise on flow estimates. Most techniques of the estimate of optical flow
have issues with aries when spatial smoothness is assumed to be broken. Nagel and Enkelmann
have established oriented smoothness to avoid fluttering of optical flow across picture borders by
observing that flow borders typically correspond with image borders. The Nagel-Enkelmann
technique was modified by Alvarez et al. such that less smoothing is done around picture
borders. The degree of smoothing has been heuristically calculated along and across borders. In
the framework of the patch-based motion discontinuity model Fleet et al. learned a statistical
model concerning picture edge orientation and amplitude to the flow limits. Black presented an
MRF model, which linked edges with edges in brightness pictures in the flow field. But this
model was crafted and tweaked by hand. We propose a probabilistic framework to learn from
instances the parameters of a model such as Nagel and Enkelmann.
Methods and Results:
Discussion and Conclusion:
We examined optical flow statistics and brightness constant statistics and built a well-
learned probabilistic model for the estimate of optical flow using a library of picture sequences
containing ground truth optical flow fields. In order to simulate the consistency of linear filter
answers, we expanded our original formula by modelling the directed derivatives of optical flow
and generalised the data term. This gave statistical basis and extension to several prior optical
flow models and at the same time allowed us to automatically learn all model parameters from
training data. A few training sequences with ground truth flow are already accessible. The basic
goal of the studied flow model is a complete general training set; it is also feasible to use
particular purpose models, of course. Although a short course set can reduce the performance of
a learning flow model, the training of the model parameters (especially individual sequences) is
preferred to a hand tuning method that is dominating. The optimising technique can also play a
major part when focusing on the goal function, and certain models can allow for better methods
of optimization than others. In addition to improved optimisation, future work may take into
account modelling the steered flow model through a picture gradient force, learning a model that
adds spatial integration to the proposed filter response constraints and therefore extends the
studied filter model over two frames, adapting the model automatically to each sequence
properties and learning an explicitly learnt filter model.
Bibliography:
1. Gennert, M.A., Negahdaripour, S.: Relaxing the brightness constancy assumption
in computing optical flow. Technical report, Cambridge, MA, USA (1987)
2. Haussecker, H., Fleet, D.: Computing optical flow with physical models of bright-
ness variation. IEEE TPAMI 23, 661–673 (2001)
3. Toth, D., Aach, T., Metzler, V.: Illumination-invariant change detection. In: 4th
IEEE Southwest Symposium on Image Analysis and Interpretation, pp. 3–7 (2000)
4. Adelson, E.H., Anderson, C.H., Bergen, J.R., Burt, P.J., Ogden, J.M.: Pyramid
methods in image processing. RCA Engineer 29, 33–41 (1984)
5. Alvarez, L., Deriche, R., Papadopoulo, T., Sanchez, J.: Symmetrical dense optical
flow estimation with occlusions detection. IJCV 75, 371–385 (2007)