---

55:148 Digital Image Processing
55:247 Image Analysis and Understanding

Chapter 14, Motion Analysis:

· Interest in motion processing has increased with advances in motion analysis methodology and processing capabilities.

· Input - a temporal image sequence

· Motion analysis often connected with real-time analysis, for example, for robot navigation.

· Common motion analysis problems

o obtain comprehensive information about moving and static objects in a scene.

o Detecting 3D shape and relative depth from motion

· Assumptions - prior knowledge helps to decrease the complexity of analysis.

o information about the camera - mobile or static

o information about time interval between consecutive images

o does the sequence represent continuous motion.

· No foolproof technique in motion analysis, no general algorithm

· Three main groups of motion-related problems

· Motion detection

o simplest, registers any detected motion

o security

o single static camera

· Moving object detection and location

o Camera usually in a static location and objects are moving

o or camera moves and objects are static

o considerably more difficult in comparison with the first group.

o Moving object detection only - solution based on motion-based segmentation methods

o More complex problems include detection of a moving object, trajectory of its motion, prediction of its future trajectory

o Image object-matching techniques often used

o direct matching

o matching of object features,

o matching of specific representative object points (corners, etc.)

o representing moving objects as graphs and matching of these graphs

o Practical examples

§ cloud tracking from a sequence of satellite meteorological data

§ cloud character and motion prediction,

§ motion analysis for autonomous road vehicles,

§ automatic satellite location by detecting specific points of interest on the

§ Earth's surface

§ city traffic analysis

§ many military applications.

§ Most complex methods work even if both camera and objects are moving.

· Derivation of 3D object properties from 2D projections acquired at different time instants of object motion - 3D object reconstruction problem

· Motion analysis = dynamic image analysis - frequently based on a small number of consecutive images

· Similar to analysis of static images, motion analyzed at a higher level, looking for correspondence between pairs of points of interest in sequential images.

· Extensive application of matching

· 2D representation of 3D motion - motion field

o each point assigned a velocity vector corresponding to

§ motion direction

§ velocity

§ distance from an observer

· Motion from optical flow

o requires a very small time distance between consecutive images

o no significant changes between two consecutive images

§ motion direction

§ motion velocity

o Aim of optical flow-based image analysis - determine a motion field.

o Illumination changes reflected in optical flow

o Object motion parameters derived from optical flow vectors

o Estimates of optical flow or point correspondence are noisy

o 3D interpretation of motion is ill-conditioned

o Requires high precision of optical flow or point correspondence.

---

· Object assumptions can help localize moving objects …

· Maximum velocity

· Small acceleration

· Common motion

· Mutual correspondence

---

Differential motion analysis methods

· Difference image d(i,j) - binary image

· non-zero values - areas with motion = difference between gray-levels in consecutive images f_1 and f_2:

· epsilon is a small positive number

· f_1, f_2 - two consecutive images separated by a time interval.

· d(i,j) - difference image has value one if:

o f_1(i,j) is a pixel on a moving object

o f_2(i,j) is a pixel on the static background

o or vice versa

o f_1(i,j) is a pixel on a moving object

o f_2(i,j) is a pixel on another moving object

o f_1(i,j) is a pixel on a moving object

o f_2(i,j) is a pixel on a different part of the same moving object

o Noise, inaccuracies of stationary camera positioning, etc.

· Results highly dependent on object--background contrast

· Trajectories detected using differential image motion analysis may not reveal the direction of the motion.

· If direction is needed, cumulative difference image may help

o constructed from a sequence of n images, with first image f_1 being a reference image

o how often

· by how much the image gray-level is different from the gray-level of the reference image

· weight coefficients a_k - significance of images in the sequence:

· more recent images may be given greater weights to reflect the importance of current motion

· A problem may be the impossibility of a static reference scene if the motion never ends

· a learning stage must construct the reference image

· Motion trajectories

o often only the center of gravity trajectory needed

· Task is simplified if objects can be segmented in the first image of sequence.

· Difference image carries information about presence of motion, characteristics of motion derived from it are not very reliable. The motion

· Robustness can be improved if intensity characteristics are compared.

· Robust motion detection - comparea corresponding areas of the image

· Detecting moving edges

· detection of slow-moving edges and weak edges

· Moving edges can be determined by logical AND operations of the spatial and temporal image edges

· S(i,j) - edge magnitudes in one of the two frames

· D(i,j) - absolute difference image

---

Last Modified: 2003