from: "Proceedings of the International Meteor Conference 1994" (1995), p.51

M O V I E - Analysis of Video Meteors


On the IMC '93 I presented main features of meteor observation with video systems and introduced our own video system called MOVIE (for details see [1]). Since no further equipment improvements had to be done, it was now time for the development of a software package for the analysis of video tapes.
As described last year, there are two main tasks, that a computer can solve in this area:

Since the number of unanalysed video tapes became bigger and bigger, I focused my work on the second point, leaving the meteor search as a task for the future.
In the last years several people published already papers with algorithms for the determination of video meteor parameters in the IMC Proceedings ([2],[3]). In the same time the power and availability of cheap PC's and their peripheral hardware increased significantly. This is very important, because fast digital image processing and analysis generally needs a powerful hardware. Furthermore you deal with huge numbers of meteors compared to meteor photography, so efficiency and time consumption are important factors in the analysis work. The more automatic and fast the computer is able to determine the meteor data, the more meteors can be analysed and (for many cases) the more accurate your data will become. So the computer should nowadays solve problems automatically, that were impossible only a few years ago.
In the first part of this paper I describe the algorithms that are used to derive the main meteor parameters like coordinates, brightness, time and speed from the video tapes. To get an idea of the problems connected with that, look at Figure 1a and 1b. Here you see both one of eight noisy, distorted and rotated single video frames of a meteor and the final meteor image after some special image processing procedures.

Figure 1a: a single video frame
Figure 1b: the final meteor image

[Figure 1a] [Figure 1b]

The second part of this paper deals with possible investigations on the basis of the analysed video meteors. Here you find also some first results of our analysis work in the last months.

a.) Computation of Meteor Parameters

Meteor Coordinates

The basis for the coordinate calculation are a noise free background image containing stars up to the 6th magnitude and the start and end video frame of the meteor. The background image is derived by averaging 10 successive video frames and subtracting the sky brightness (which is not constant in the field of view). The other two frames are the first and the last frame from the complete digitised series of meteor images (see [2] will not occur, because all shadows are completely removed by the sky brightness subtraction at the beginning of the analysis.
The accurate determination of the linear meteor position in the single video frames is not that easy. The signal to noise ratio of such a frame is generally much worse than for of the essentially noise free background image, and bright meteors often show long tails. So you cannot calculate the centre of the meteor image, but have to point a cross on the meteor's centre in an enlarged image 'by hand'. Hence the achieved accuracy is only less than or about one pixel.
At this point linear positions of the objects in the image are known, but they have to be transformed into equatorial coordinates. To do that, the user first has to identify at least 3 reference stars in the field of view manually. This can be done very comfortable: All automatically recognised stars flash cyclic one after another on the screen and you select known stars from a prepared list of about 150 bright stars. These stars are the basis for a first plate constant calculation, after which the rough equatorial coordinates of all objects in the image are known. In a second step the program searches automatically in a star catalogue ( I use the PPM catalogue reduced to stars up to 7 mag) close to the calculated positions for stars, that have about the same brightness as the unidentified objects in the image. Now stars in the whole field of view are recognised and the following second plate constant calculation gives more accurate equatorial coordinates. In the last step all objects with large position errors are removed from the list of reference stars to avoid misidentifications, and a last time the plate constants are calculated.
If you use a guided video system this comfortable procedure becomes even more simple for you, because you have to identify the first three reference stars only once. The computer stores their position and recognises them automatically during all following coordinate calculations.
After the plate constant calculation you know both the linear meteor position and the transformation function between linear and equatorial coordinates (described by the 'plate constants'). So you can finally calculate the right ascension and declination of your meteor.
The last open question is the algorithm used for the determination of the plate constants. There are several approaches of different complexity published in the papers: In [2] a local linear fitting using only stars in the neighbourhood of the meteor is proposed, in [3] higher order terms for the global fitting function are used as well. For some reasons I used still another method of first 'undistorting' the image and then using a global linear fit for the plate constants.
Figure 2a shows an optical test wall at the Technical University of Berlin, recorded with our MOVIE system in its standard configuration (see [1] for details). The distortion of the quadrates is well visible, it increases rapidly with larger distances from the centre of the field of view. The analysis of this image showed, that the distortion can be accurately described by an exponential function (Figure 3) similar to the distortion functions of fisheye photographs. Using the inverse of this function you cannot only compute the undistorted linear positions of all objects, but also generate a distortion-free image with the methods of digital image processing. Figure 2b shows the result of applying this procedure to the test image. As this figure shows, almost all distortion disappeared. The remaining deformation is of an unsystematic nature and will be considered by the plate constants. They are linear fitted, using stars in the whole image. In the future I will use local fitting as described in [2] as well, because it seems to increase the resulting accuracy by about a factor of 2.

Figure 2a: a distorted test image
Figure 2b: after removing the distortion

[Figure 2a] [Figure 2b]
Figure 3:

[Figure 3]

Resulting from the coordinate calculation you have the equatorial meteor positions, which are stored in a database. For compatibility reasons the described program uses the PosDat format, so other software packages like Radiant can be used right after the analysis.

Meteor Brightness

Stars and the meteor are not single pixels but bloomed spots in the image. So you cannot look only at the brightest pixel but have to take all pixels belonging to an object into consideration. If you add all the pixel values and take the logarithm of the sum, you get an accurate measurement for the apparent brightness of the object. Amazingly this works even at the edges of the field of view quite well. During the computation of the centre of a star image the program checks already all pixels of an object. So you can already add their values there and store the sum together with the mean position. The magnitudes of the identified reference stars is known from the star catalogue. Hence the program only has to calibrate the brightness scale from the pixel sums with the help of these reference brightness values, and you know the meteor brightness.
The algorithm works very good in a wide range of cases. Significant errors will only occur for very faint (due to the major influence of noise) and very bright (due to the limited maximal brightness in a digital image) objects. In the first case you sometimes have to estimate the brightness visually, in the second case more sophisticated algorithms like estimations of the maximal brightness from the pixel value slope may be implemented in the future.

Time, Duration and Speed of Meteors

In contrast to meteor photography, all these parameters are quite easy to determine. For the accurate time you use the clock of the Camcorder, which will be superimposed to the lower right corner of your video image. The duration can be determined by simply counting the number of video frames, that contain the meteor. Here you use the fact, that the PAL video standard defines 50 interlaced video frames per second with a fixed time of 0.02s between two frames. Finally you can calculate the meteor speed from the angular distance between the meteors start and end point and its duration.

Computation of a Meteor Image

It is not necessary for the determination of the meteor parameters to generate a nice meteor picture from the single video frames, but a video meteor consisting only of a reference number and some parameters in a database is quite boring. So for each analysed meteor the program superimposes the complete meteor trail on the noiseless background image, stretches the grey scale and increases the contrast, removes the distortion, turns it into the right direction and finally converts it into the JPEG picture format. You finish up with a picture, that tells much more than only the pure numbers, but needs almost no disc space due to the good compression algorithm of JPEG.


After applying all the mentioned steps you have generated a new entry in your PosDat database, which could look as follows:

 173       8     21   2   7   1.2 26   10   301.5  53.4   295.1  47.3   2 AAJ 0.28s
Furthermore you have a 4 kByte long file with the name 'AAJ-008.JPG' on your hard disc, which contains the image of Figure 1b.
Using the actual version of the analysis program and MOVIE in its standard configuration (field of view: 60 degrees), you get accuracy's of about 10' in the meteor and star positions. That is nearby the size of one pixel in the distorted image. Even if the star positions will become better due to improvements in the coordinate calculation algorithm, the meteor position actually cannot be improved much due to the described 'by hand' selection of the meteor position. The meteor brightness is more accurate than 0.3 mag, only for meteors fainter than 4 mag or brighter than -2 mag larger errors may occur. The meteor speed accuracy depends from the accuracy of the meteor position, but the error will only for very short meteors exceed 2 degrees per second. The accuracy of the meteor time depends on how accurate you have adjusted your Camcorder clock. You need in average about 5 minutes to digitise and analyse one meteor, which is a quite reasonable result. On the other hand it still takes you hours and hours, if you want to analyse complete tapes, that were re- corded near the maximum of active meteor showers.

b.) Investigations with the Derived Data

Once you have filled your database with a certain amount of accurate video meteor data, you can start to do a lot of different analysis. In the following I will give some ideas of possible investigations and present first results from the analysis of our '93 and '94 Perseid video tapes.

ZHR calculations

You can use the fact, that a video system like our MOVIE is almost as effective but much more objective than a visual observer, to determine ZHR's of both minor and major meteor showers. Therefore you first of all do not need all the meteor data described above, but a list of all shower meteors and their rough times of appearance. Just this is the work of the automatic meteor search system, which is not yet implemented. On the other hand we had to watch the tapes visually to make any analysis possible, so I can present a first ZHR plot from the Perseid maximum '93 here.
Of course you have to think more detailed about how you have to change the ZHR calculation procedure to make it suitable for video observations, if you want to compute accurate ZHR's. In our first test we used the standard formula for visual observations bearing in mind, that the probability of finding a meteor visually on the tapes is similar to the one recognising the meteor in the night sky. Furthermore we can assume, that our MOVIE is in contrary to the video systems of other groups [2,4,5] much more similar to visual than to telescopic observations due to the wide angle foto lense we use. The limiting magnitude of the video system was determined as for normal visual observation by counting stars in special reference areas.
Figure 4 shows the resulting ZHR curve of MOVIE on August 11/12, 1993, compared with our parallel obtained visual rates and the 'official' IMO ZHR's. There is a correlation between the curves, but especially near the maximum in the morning twilight MOVIE became much more effective than any visual observer. While the visual rates were about twice the video rates during the whole night (which is to expect from MOVIE's limited field of view of only 60 degrees), they became significant smaller in the last two periods. It is possible, that video systems do not suffer as strong as visual observer from a bright sky, which was our expression in 1993 as well, but we have to do further test to prove this result.

Figure 4: ZHR for video and visual observations

[Figure 4]

Meteor clustering

This is a topic, that is discussed again and again, but not finally decided yet due to missing accurate and objective observations. In another paper the author showed, that the personal impression of a visual observer is a bad measurement for real clustering [6]. With video systems we are now able to search for clusters still better than with our computer based visual meteor observation (see [7] for details), because even shortest time intervals of a fraction of a second between meteors can be measured accurately. This was impressive shown in [8], although the Canadian observers noticed clustering at a much smaller time scale than we intend to do. Up to now we have not yet looked at our data from that point of view.

Test of Meteor Brightness Estimations

Video meteor analysis enables us not only to calculate the mean brightness of a meteor, but the complete brightness development along the meteor trail. That makes it possible to compare visual brightness estimations of meteors with the video data. From the analysis of the Perseids '94 we can show, that the visual observers estimated the meteors in average one magnitude to faint compared to the maximal video meteor brightness. Figure 5 was derived from the analysis of 68 parallel meteor observations, corresponding to 125 visual brightness estimations. The first graph shows the meteor brightness distribution for both video and visual observations. The displacement of the curves is well visible. The second graph gives the distribution of the direct differences between the magnitudes. Here you have to take into consideration, that visual brightness estimations were done only in steps of 1 mag, whereas the video meteor brightness is calculated with an accuracy of 0.1 mag.

Figure 5: video and visual meteor brightness

[Figure 5a]

errors of visual brightness estimations

[Figure 5b]

Calculation of Meteor Shower Radiants

Using the well-known Radiant software it is relatively easy to calculate from the given database actual positions of known meteor showers and search for unknown ones. Figure 6 shows the radiant plot for the Perseids, which was derived from the coordinates of 109 meteors observed during our meteor expedition '94.

Figure 6: radiant plot of the Perseids '94

[Figure 6]

Once you know, which meteor belongs to a certain meteor shower, impressive pictures like Figure 7 are easy to compute from your database of distortion-free meteor images. The image shows the brightest Perseids recorded on August 09/10, 12/13 and 13/14 in Krampfer/Germany.

Figure 7: Perseids in the Summer triangle

[Figure 7a]

Meteor Parallaxes

The use of two similar video systems enables you to observe a lot of meteors from two parallel stations and therefore to calculate hundreds of meteor trajectories. The software changes are minimal. You only have to calculate all the meteor positions in the trail instead of the first and last point. For the meteor in Figure 1b this additional information looks as follows:

date                        : 13/14.08.94
time of the meteor          : 22:02:07.76 -  22:02:08.04 UT
duration                    : 0.28 s
starting point              : Alpha = 301.5 deg, Delta =  53.4 deg
end point                   : Alpha = 295.1 deg, Delta =  47.3 deg
number of reference stars   : 34
mean position error         : 9.9'
mean velocity               : 26 degrees/s
maximal brightness          :  1.2 mag
single positions            :
22:31:07.76 UT              : Alpha = 301.5 deg,   Delta =  53.4 deg,  2.8 mag
22:31:07.80 UT              : Alpha = 300.7 deg,   Delta =  52.5 deg,  1.9 mag
22:31:07.84 UT              : Alpha = 299.7 deg,   Delta =  51.7 deg,  1.6 mag
22:31:07.88 UT              : Alpha = 298.8 deg,   Delta =  50.9 deg,  1.2 mag
22:31:07.92 UT              : Alpha = 297.7 deg,   Delta =  49.8 deg,  1.2 mag
22:31:07.96 UT              : Alpha = 296.8 deg,   Delta =  49.1 deg,  1.3 mag
22:31:08.00 UT              : Alpha = 296.1 deg,   Delta =  48.3 deg,  1.6 mag
22:31:08.04 UT              : Alpha = 295.1 deg,   Delta =  47.3 deg,  2.3 mag

This year we did not obtain any double observations of video meteors together with our Dutch friends from the NVWS, mainly because of the bad weather conditions and some problems with the second video system. On the other hand we have planned already the next parallel video observation session and hope, that we can present first results in the next year.

Meteor Spectra

It is quite a hard work to obtain spectra of meteor trails with photo cameras, because you need very bright meteors and a fast camera operator (see [9] for an example). With the help of video equipment in this completely different area of research you should be able to record spectra of much fainter meteors automatically, which will increase the efficiency of the recording system by at least a factor of ten.


In the last months the author developed a software package, that allows the efficient and accurate analysis of video meteors. We used it to analyse our recordings of the Perseids in 1994. On the basis of the digitised and analysed 173 meteors we started to do further investigation like radiant calculations and brightness estimation checks, but the variety of analysis is large and we are by far not able to do all the possible investigation alone. All data are stored in PosDat format and available from the IMO database, so everybody can use it for his own purposes.
The software has still prototype function. Many routines are hardware dependent and not easy to adapt to other computer systems. After further improvements and reliability checks it will be rewritten in a more independent type to make it available for other users of video systems.
The program for the automatic meteor search will be implemented in the future as well.


I want to thank the Archenhold-Observatory Berlin for the provision of the image intensifier, which made all the video work possible. Thanks to Arno Gnädig, who implemented the plate constant computation, and Rainer Arlt, who generated the radiant plot. Finally special thanks to the members of our MOVIE team for their help and ideas: Kathrin Düber, Mirko Nitschke and David Przewozny.


[1] Molau, S. (1993), "MOVIE - Meteor Observation with VIdeo Equipment",
Proceedings of the International Meteor Conference 1993, p.71

[2] Hawkes, R.L. et al. (1992), "Analysis Procedures for Two Station Television Meteors",
Proceedings of the International Meteor Conference 1992, p.28

[3] De Lignie, M. and Jobse, K. (1989), "Accurate Radiant Determination from TV Meteors",
Proceedings of the International Meteor Conference 1989, p.41

[4] Ueda, M. et al. (1993), "TV Observations of the 1991 Geminid Meteor Shower",
WGN - Journal of the International Meteor Organization 21-4, p.215

[5] Suzuki, S. et al. (1994), "Multi-Station TV Observations of the 1993 Perseids",
WGN - Journal of the International Meteor Organization 22-4, p.137

[6] Molau, S. (1992), "Murphy's Effekt bei der Meteorbeobachtung?",
Mitteilungen des Arbeitskreises Meteore 140, p.5

[7] Nitschke, M. (1991), "Computer-Based Meteor Observation",
Proceedings of the International Meteor Conference 1991, p.54

[8] Piers, P.A. et al. (1993), "An Unusual Meteor Cluster Observed by Image-Intensified Video",
WGN - Journal of the International Meteor Organization 21-4, p.168

[9] Zimnikoval, P. et al. (1993), "A New Camera for Registration of Persistent Trains",
Proceedings of the International Meteor Conference 1993, p.85-87

Sirko Molau; last change: July 18, 1996