GB2365244A

GB2365244A - Image stabilisation

Info

Publication number: GB2365244A
Application number: GB0018501A
Authority: GB
Inventors: Mark Lebbell; David Tasker
Original assignee: Snell and Wilcox Ltd
Current assignee: Snell Advanced Media Ltd
Priority date: 2000-07-27
Filing date: 2000-07-27
Publication date: 2002-02-13
Also published as: GB0018501D0

Abstract

In processing a video image from a camera mounted on an unstable platform, the image is shifted in response to a global motion vector so as to enable an observer to gain the maximum information about an object in the picture. Respective limits are set to the amount of horizontal and vertical picture shift that can be applied, and when a correction limit is reached, an abrupt change of the respective correction is made to a preset value within the relevant limits.

Description

<Desc/Clms Page number 1> IMAGE STABILISATION This invention concerns the processing of video images from a camera on an unstable platform (such as a mast or a boat) so remove the effects of camera unsteadiness.

UK patent application 0012349.7 describes a method of improving the subjective quality of a picture from a video camera on an unstable mounting; motion vectors and associated confidence values are processed to obtain a global motion vector which is used to shift the picture so as to cancel out the camera unsteadiness. Where the resulting video signal is to be used in broadcasting and similar media for information, education or entertainment, it is important that the processing is imperceptible to the viewer and the aesthetic quality of the picture is not impaired in any way. For this reason care is taken to allow deliberate camera pans and tilts to be portrayed properly, and the picture is enlarged slightly so that the edges of the original picture are placed outside the active picture area so that they do not reveal the action of the unsteadiness corrector.

The inventors have appreciated that there is another application for the sort of unsteadiness correction described in the above-mentioned patent which is the processing of a video image from a camera on an unstable platform so as to enable an observer to gain the maximum information about an object in the picture. In this application the aesthetic quality of the picture is of no concern, the camera is trained on a particular object of interest and deliberate pans or tilts do not occur.

The invention consists in one aspect of a method of processing a video image from a camera on an unstable platform by shifting the image in response to a global motion vector so as to enable an observer to gain the maximum information about an object in the picture, characterised in that respective limits are set to the amount of horizontal and vertical picture shift that can be applied and when a correction limit is reached an abrupt change of the respective correction is made to a preset value within the relevant limits.

Suitably, the preset value is substantially at the centre of the relevant correction range when alternating motion is detected; and, the preset value is substantially at the opposite end of the relevant correction range when steady motion is detected.

In a further aspect the invention consists of a method of processing a video image from a camera on an unstable platform by shifting the image in response to a global motion vector so as to enable an observer to gain the maximum information about an object in the picture, characterised in that the blanking edges of the input picture appear within the active area of the output picture as the amount of applied picture shift increases.

Advantageously, the amount of picture shift applied may exceed 10% of the smallest picture dimension.

An example of the invention will now be described with reference to the drawings in which:- Figure 1 shows the block diagram of a known unsteadiness corrector; and Figure 2 shows the block diagram of an unsteadiness corrector in accordance with the invention.

Referring to Figure 1, an input video signal (1) is accompanied by a global motion vector signal (2). The global motion vector signal is derived in known manner, possibly as described in UK patent application 0012349.7.

A resolver (3) separates the vectors into horizontal and vertical components, which are high-pass filtered in respective filters (4) and (5). The filters remove the DC and low frequency motion due to deliberate camera movements and allow high frequency motion due to camera unsteadiness to pass with little attenuation. These filtered vectors are integrated in respective horizontal and vertical integrators (6) and (7) to generate horizontal and vertical displacement signals (8) and (9), which will be used to correct the picture for unsteadiness.

The input video signal (1) is delayed in a delay stage (13), which compensates for the time taken to process the motion vectors, and input to a picture shifter (12). This is able to move the picture in response to horizontal and vertical displacement signals (14) and (15). The picture is enlarged slightly prior to shifting so that although information at the edges falls outside the active picture area and is blanked, limited displacement of the picture in any direction is possible without any picture edge appearing to move.

The displacement signals (8) and (9) are limited in non-linear circuits (10) and (11) so as to ensure that the picture shift is never large enough to bring the original blanking edges into the active picture area.

The output (16) of the shifter (12) therefore shows a steady picture in which slow camera movements remain and the picture content always fills the active picture area.

Figure 2 shows an improved system for correcting unsteadiness in accordance with the invention. (Some elements of this figure correspond with elements of Figure 1 and these have the same identification numerals prefixed by the digit 2). The processing of the global motion vectors differs from that shown in Figure 1 so as to obtain an output video signal more suitable for surveillance purposes.

The input video signal (201) is accompanied by a global motion vector signal (202), which is resolved (203) into horizontal and vertical components in the same way as described previously. However, the vector processing does not include the high-pass filtering and amplitude limiting stages of the system shown in Figure 1; and, the picture shifter (212) need not necessarily magnify the picture before shifting it, although it may sometimes be desirable to apply a very high degree of magnification where a small, or distant object is to be observed. The amount of shift which can be provided is much larger than in the prior art system and may be of the order of one quarter of the picture width or height if no magnification is applied, or may approach the maximum picture dimension if a large magnification is applied.

Referring to Figure 2, the processing (220) of the horizontal component of the global vector signal from the resolver (203) will now be described. The vector signal is integrated (206) so as to obtain a displacement signal which controls the horizontal shift which is applied in the shifter (212). The integrator (206) can be reset to either the positive extremity, centre or negative extremity of its output voltage range by respective reset control signals (221), (222) and (223). These voltage limits are arranged to correspond to the maximum desirable values of picture shift to be applied in the block (212). The precise values chosen will depend of the size of the object of interest in the picture relative to the overall picture size and may conveniently be controlled by an operator.

The reset signals are generated in response to the outputs from comparators (224) and (225), which give outputs when the integrator reaches the positive and negative extremities respectively of its output range.

Changes in the sign of the horizontal component of the global motion vector (202) are detected in the block (226) and a counter (227) is incremented for every field where the sign is the same as the previous field, and reset to zero whenever the sign differs from the previous field. The output of the counter is compared (228) with a threshold (229) to give a logic signal (230) which is active when monotonic motion is detected and inactive when the motion is alternating. This monotonic-motion logic signal (230) enables the AND-gates (231) and (232) and disables the AND-gate (233) so as to route the output of the positive limit detecting comparator (224) to the negative integrator reset control (223), and the output of the output of the negative limit detecting comparator (225) to the positive integrator reset control (221). This ensures that when the maximum permitted horizontal picture shift (in either direction) is reached the integrator is immediately reset to the opposite end of its range. The picture is thus shifted as far as possible in the opposite direction to the previously applied unsteadiness correction so as to allow the position of the object of interest on the screen to be held substantially constant for the maximum length of time before the correction "runs out" and another reset is required. (The skilled person will recognise that the above arrangement is prone to undesired oscillation and will avoid this in known manner, perhaps by introducing hysteresis into the comparators.)

Where the motion is non-monotonic the logic signal (230) is inactive and the outputs from both comparators (224) and (225) are routed via the OR-gate (226) and the AND-gate (233) to the reset-to-centre control input (222) of the integrator (206). In this case the picture is centred whenever the maximum horizontal shift is reached in either direction, thus allowing unsteadiness correction in either direction immediately after the reset operation.

The vertical component of the global motion vector signal (202) from the resolver (203) is processed in the block (221) in the same way as described for the horizontal component in the block (220). The resulting vertical picture shift signal (215) controls the picture shifter (212) together with the horizontal picture shift signal (214) to give a corrected video output signal (216).

It must be recognised that the invention has been described only by way of example and many variations are possible within the concepts described. For example alternative methods of distinguishing between monotonic and alternating motion may be employed. Another example would be a system where the input is switched between several cameras, and the horizontal and vertical integrators are reset when the source is switched, possibly under the control of a "shot- change" detector.

Claims

CLAIMS 1. A method of processing a video image from a camera on an unstable platform by shifting the image in response to a global motion vector so as to enable an observer to gain the maximum information about an object it the picture, characterised in that respective limits are set to the amount of horizontal and vertical picture shift that can be applied and when a correction limit is reached an abrupt change of the respective correction is made to a preset value within the relevant limits.
2. A method according to Claim 1, in which the preset value is substantially at the centre of the relevant correction range when alternating motion is detected.
3. A method according to Claim 1, in which the preset value is substantially at the opposite end of the relevant correction range when steady motion is detected.
4. A method of processing a video image from a camera on an unstable platform by shifting the image in response to a global motion vector so as to enable an observer to gain the maximum information about an object in the picture, characterised in that the blanking edges of the input picture appear within the active area of the output picture as the amount of applied picture shift increases.
5. A method of processing a video image from a camera on an unstable platform by shifting the image in response to a global motion vector so as to enable an observer to gain the maximum information about an object in the picture, characterised in that the amount of allowable picture shift exceeds 10% of the smallest picture dimension.