3.1. Detection Parameters

The meteors detection chain is located here: ./bin/fmdt-detect.

The following table summarizes the available parameters:

Argument

Type

Details

--vid-in-path

STRING

See Section 3.1.2.

--vid-in-start

INTEGER

See Section 3.1.3.

--vid-in-stop

INTEGER

See Section 3.1.4.

--vid-in-skip

INTEGER

See Section 3.1.5.

--vid-in-buff

BOOLEAN

See Section 3.1.6.

--vid-in-loop

INTEGER

See Section 3.1.7.

--vid-in-threads

INTEGER

See Section 3.1.8.

--vid-in-dec

STRING

See Section 3.1.9.

--ccl-impl

STRING

See Section 3.1.10.

--ccl-hyst-lo

INTEGER

See Section 3.1.11.

--ccl-hyst-hi

INTEGER

See Section 3.1.12.

--ccl-fra-path

STRING

See Section 3.1.13.

--ccl-fra-id

BOOLEAN

See Section 3.1.14.

--cca-mag

BOOLEAN

See Section 3.1.15.

--cca-ell

BOOLEAN

See Section 3.1.16.

--cca-roi-max1

INTEGER

See Section 3.1.17.

--cca-roi-max2

INTEGER

See Section 3.1.18.

--mrp-s-min

INTEGER

See Section 3.1.19.

--mrp-s-max

INTEGER

See Section 3.1.20.

--knn-k

INTEGER

See Section 3.1.21.

--knn-d

INTEGER

See Section 3.1.22.

--knn-s

FLOAT

See Section 3.1.23.

--trk-ext-d

INTEGER

See Section 3.1.24.

--trk-ext-o

INTEGER

See Section 3.1.25.

--trk-angle

FLOAT

See Section 3.1.26.

--trk-star-min

INTEGER

See Section 3.1.27.

--trk-meteor-min

INTEGER

See Section 3.1.28.

--trk-meteor-max

INTEGER

See Section 3.1.29.

--trk-ddev

FLOAT

See Section 3.1.30.

--trk-ell-min

FLOAT

See Section 3.1.31.

--trk-all

BOOLEAN

See Section 3.1.32.

--trk-roi-path

STRING

See Section 3.1.33.

--log-path

STRING

See Section 3.1.34.

--vid-out-path

STRING

See Section 3.1.35.

--vid-out-play

BOOLEAN

See Section 3.1.36.

--vid-out-id

BOOLEAN

See Section 3.1.37.

3.1.1. Standard Output

fmdt-detect outputs a list of tracks. The tracks represent the detected objects in the video sequence. Here is the template of the output text:

# -------||---------------------------||---------------------------||---------
#  Track ||           Begin           ||            End            ||  Object
# -------||---------------------------||---------------------------||---------
# -------||---------|--------|--------||---------|--------|--------||---------
#     Id || Frame # |      x |      y || Frame # |      x |      y ||    Type
# -------||---------|--------|--------||---------|--------|--------||---------
   {tid} ||  {fbeg} | {xbeg} | {ybeg} ||  {fend} | {xend} | {yend} || {otype}

  • {tid}: a positive integer (start from 1) value representing a unique track identifier,

  • {fbeg}: a positive integer value representing the first frame in the video sequence when the track is detected,

  • {xbeg}: a positive real value of the x-axis coordinate (beginning of the track),

  • {ybeg}: a positive real value of the y-axis coordinate (beginning of the track),

  • {fend}: a positive integer value representing the last frame in the video sequence when the track is detected,

  • {xend}: a positive real value of the x-axis coordinate (end of the track),

  • {yend}: a positive real value of the y-axis coordinate (end of the track),

  • {otype}: a string of the object type, can be: meteor, star or noise.

3.1.2. --vid-in-path

Deprecated

--in-video

Type

STRING

Default

[empty]

Example

--vid-in-path ~/Videos/meteors.mp4

Input video path (supports also a path to a sequence of images path/basename_%05d.jpg).

3.1.3. --vid-in-start

Deprecated

--fra-start

Type

INTEGER

Default

0

Example

--vid-in-start 12

First frame id (included) to start the detection in the video sequence.

3.1.4. --vid-in-stop

Deprecated

--fra-end

Type

INTEGER

Default

0

Example

--vid-in-stop 42

Last frame id (included) to stop the detection in the video sequence. If set to 0, read entire video.

3.1.5. --vid-in-skip

Deprecated

--fra-skip

Type

INTEGER

Default

0

Example

--vid-in-skip 1

Number of frames to skip.

3.1.6. --vid-in-buff

Deprecated

--video-buff

Type

BOOLEAN

Example

--vid-in-buff

Bufferize all the video in global memory before executing the chain.

3.1.7. --vid-in-loop

Deprecated

--video-loop

Type

INTEGER

Default

1

Example

--vid-in-loop 10

Number of times the video is read in loop.

3.1.8. --vid-in-threads

Deprecated

--ffmpeg-threads

Type

INTEGER

Default

0

Example

--vid-in-threads 1

Select the number of threads to use to decode video input (in ffmpeg). If set to 0, ffmpeg chooses the number of threads automatically.

3.1.9. --vid-in-dec

Type

STRING

Default

FFMPEG-IO

Example

--vid-in-dec VCODECS-IO

Select the input video decoder interface. FFMPEG-IO is based on the cmd line ffmpeg executable to exchange decoded frames over a system pipe. VCODECS-IO directly call the libav.

Note

VCODECS-IO works only if FMDT has been compiled with the CMake -DFMDT_USE_VCIO=ON option (see Section 2.2.1).

3.1.10. --ccl-impl

Type

STRING

Default

LSLH

Example

--ccl-impl LSLH

Choose the LSL implementation. Can be LSLH or LSLM.

LSLH is the implementation discribed in [LZ09] and LSLM is the implementation discribed in [LHL20].

Note

LSLM is only available if FMDT has been compiled with the CMake -DFMDT_LSL_LINK=ON option (see Section 2.2.1).

3.1.11. --ccl-hyst-lo

Deprecated

--light-min

Type

INTEGER

Default

55

Example

--ccl-hyst-lo 100

Minimum light intensity for hysteresis threshold (grayscale \([0;255]\)).

3.1.12. --ccl-hyst-hi

Deprecated

--light-max

Type

INTEGER

Default

80

Example

--ccl-hyst-hi 140

Maximum light intensity for hysteresis threshold (grayscale \([0;255]\)).

3.1.13. --ccl-fra-path

Deprecated

--out-frames

Type

STRING

Default

[empty]

Example

--ccl-fra-path ccl_fra/%05d.png

Path of the files for CC debug (path/cc_%05d.png).

3.1.14. --ccl-fra-id

Deprecated

--show-id

Type

BOOLEAN

Example

--ccl-fra-id

Show the RoI/CC ids on the output frames (to combine with --ccl-fra-path parameter). Requires to link with OpenCV library (-DFMDT_OPENCV_LINK CMake option, see Section 2.2.1).

3.1.15. --cca-mag

Type

BOOLEAN

Default

[empty]

Example

--cca-mag

Enable the computation of two news features in the CCA: the magnitude and the counter of saturated pixels (to be combined with the --log-path option).

3.1.16. --cca-ell

Type

BOOLEAN

Default

[empty]

Example

--cca-ell

Enable the computation of two news features in the CCA: a the semi-major axis of an ellipse and b the semi-minor axis of an ellipse. This option has to be combined with the --log-path option.

3.1.17. --cca-roi-max1

Type

INTEGER

Default

65535

Example

--cca-roi-max1 10000

Maximum number of RoIs before hysteresis threshold. Allow to manage the memory footprint of the program. The Smaller the maximum number of RoIs, the smaller the memory footprint.

3.1.18. --cca-roi-max2

Type

INTEGER

Default

400

Example

--cca-roi-max2 200

Maximum number of RoIs after hysteresis threshold. Allow to manage the memory footprint of the program. The Smaller the maximum number of RoIs, the smaller the memory footprint.

3.1.19. --mrp-s-min

Deprecated

--surface-min

Type

INTEGER

Default

3

Example

--mrp-s-min 5

Minimum surface of the CCs in pixels.

3.1.20. --mrp-s-max

Deprecated

--surface-max

Type

INTEGER

Default

1000

Example

--mrp-s-max 50

Maximum surface of the CCs in pixels.

3.1.21. --knn-k

Deprecated

-k

Type

INTEGER

Default

3

Example

--knn-k 5

Maximum number of neighbors considered in the k-NN algorithm.

3.1.22. --knn-d

Deprecated

--max-dist

Type

INTEGER

Default

10

Example

--knn-d 25

Maximum distance in pixels between two images (k-NN algorithm).

3.1.23. --knn-s

Deprecated

--min-ratio-s

Type

FLOAT

Default

0.125

Example

--knn-s 0.0

Minimum surface ratio to match two CCs in k-NN (0 matches alls, 1 matches nothing). This parameter is also used for extrapolation in the tracking.

3.1.24. --trk-ext-d

Deprecated

--r-extrapol

Type

INTEGER

Default

5

Example

--trk-ext-d 25

Search radius in pixels for CC extrapolation (piece-wise tracking).

3.1.25. --trk-ext-o

Deprecated

--extrapol-orde

Type

INTEGER

Default

3

Example

--trk-ext-o 1

Maximum number of frames to extrapolate for lost objects (linear extrapolation).

3.1.26. --trk-angle

Deprecated

--angle-max

Type

FLOAT

Default

20.0

Example

--trk-angle 35.0

Tracking max angle between two meteors at \(t-1\) and \(t\) (in degree). This is a classification criterion.

3.1.27. --trk-star-min

Deprecated

--fra-star-min

Type

INTEGER

Default

15

Example

--trk-star-min 5

Minimum number of frames required to track a star. This is a classification criterion.

3.1.28. --trk-meteor-min

Deprecated

--fra-meteor-min

Type

INTEGER

Default

3

Example

--trk-meteor-min 5

Minimum number of frames required to track a meteor. This is a classification criterion.

3.1.29. --trk-meteor-max

Deprecated

--fra-meteor-max

Type

INTEGER

Default

100

Example

--trk-meteor-max 50

Maximum number of frames required to track a meteor. This is a classification criterion.

3.1.30. --trk-ddev

Deprecated

--diff-dev

Type

FLOAT

Default

4.0

Example

--trk-ddev 5.5

Multiplication factor of the standard deviation (CC error has to be higher than \(ddev \times stddev\) to be considered in movement). This is a classification criterion.

3.1.31. --trk-ell-min

Type

FLOAT

Default

0.0

Example

--cca-ell --trk-ell-min 3.0

Minimum ellipse ratio to be considered as a meteor. This is a classification criterion. If the value is 0 then this parameter has no effect. Moreover, this parameter requires the --cca-ell parameter to work. If the latest is not set, then this parameter is ignored.

3.1.32. --trk-all

Deprecated

--track-all

Type

BOOLEAN

Example

--trk-all

By default the program only tracks meteor object type. If --trk-all is set, all object types are tracked (meteor, star or noise).

This parameter is used in the _tracking_perform() function.

3.1.33. --trk-roi-path

Type

STRING

Default

[empty]

Example

--trk-roi-path trk2roi.txt

Path to the output file containing lists of the RoI ids of the tracked objects. Each line corresponds to a track/object and here is the corresponding line format:

{tid} {otype} {rid1} {rid2} {...} {ridn}

{rid1} is the first RoI id of the track/object of {tid} id. {rid2} is the second RoI id (in the second frame where the object has been tracked). And so on, until the last RoI id {ridn}. Note that sometime the RoI id can be 0, it means that the object has been extrapolated on this frame, thus there is no RoI id for this frame.

3.1.34. --log-path

Deprecated

--out-stats

Type

STRING

Default

[empty]

Example

--log-path detect_logs/

Path of the output statistics, only required for debugging purpose.

Warning

This section targets advanced users, some knowledge about the implemented algorithms may be required!! You have been warned ;-).

fmdt-detect comes with the --log-path option to help to understand what is happening during the execution. This option enables to log internal statistics of the different algorithms used to detect meteors.

The folder contains multiple files, one per frame. For instance, the file name for the frame n°12 is: 00012.txt. Each file contains 5 different tables:

  • Table 1: list of RoIs at \(t - 1\) (result of the CCL/CCA + hysteresis algorithm at \(t - 1\)),

  • Table 2: list of RoIs at \(t\) (result of the CCL/CCA + hysteresis algorithm at \(t\)),

  • Table 3: list of associations between \(t - 1\) RoIs and \(t\) RoIs (result of the k-NN algorithm) + errors/velocities after motion estimation,

  • Table 4: motion estimation statistics between \(t - 1\) and \(t\) frame,

  • Table 5: list of tracks since the beginning of the execution (final output of the detection chain).

Note

The first log file (usally named 00000.txt) only contains the table 2. This is normal because algorithms starting from k-NN require two consecutive frames to work.

3.1.34.1. Table 1 and table 2: RoIs

# ------||----------------||---------------------------||-------------------------------------------||-------------||-----------||------------||-------------------
#   RoI ||      Track     ||        Bounding Box       ||           Surface (S in pixels)           ||   Center    || Magnitude || Saturation ||      Ellipse
# ------||----------------||---------------------------||-------------------------------------------||-------------||-----------||------------||-------------------
# ------||------|---------||------|------|------|------||-----|------|------|-------|-------|-------||------|------||-----------||------------||-----|-----|-------
#    ID ||   ID |    Type || xmin | xmax | ymin | ymax ||   S |   Sx |   Sy |   Sx2 |   Sy2 |   Sxy ||    x |    y ||        -- ||    Counter ||   a |   b | ratio
# ------||------|---------||------|------|------|------||-----|------|------|-------|-------|-------||------|------||-----------||------------||-----|-----|-------
  {rid} || {tid}| {otype} ||{xmin}|{xmax}|{ymin}|{ymax}|| {S} | {Sx} | {Sy} | {Sx2} | {Sy2} | {Sxy} || {cx} | {cy} ||     {mag} ||      {sat} || {a} | {b} |   {r}

Each line corresponds to one RoI:

  • {rid}: unique identifier for the current RoI (start from 1),

  • {tid}: unique identifier of the corresponding track (start from 1), can be, empty if no track is associated to the current RoI,

  • {otype}: type of the track object (meteor, noise or star), only if there is a track corresponding to this RoI,

  • {xmin}: minimum \(x\) position of the bounding box,

  • {xmax}: maximum \(x\) position of the bounding box,

  • {ymin}: minimum \(y\) position of the bounding box,

  • {ymax}: maximum \(y\) position of the bounding box,

  • {S}: surface (area) of the RoI in pixels,

  • {Sx}: sum of \(x\) properties,

  • {Sy}: sum of \(y\) properties,

  • {Sx2}: sum of \(x^2\) properties,

  • {Sy2}: sum of \(y^2\) properties,

  • {Sxy}: sum of \(x \times y\) properties,

  • {cx}: \(x\) center of mass,

  • {cy}: \(y\) center of mass,

  • {mag}: magnitude of the current RoI (accumulated brightness of the RoI),

  • {sat}: number of pixels that are saturated in the current RoI (a pixel \(x\) is saturated when its intensity \(i_x = 255\)),

  • {a}: semi-major axis (ellipse),

  • {b}: semi-minor axis (ellipse),

  • {r}: ratio \(a / b\).

{mag} and {sat} features are not enabled by default (and the - character is printed in the corresponding columns). To enable theses features you need to use the --cca-mag command line parameter. For more information about those features you can refer to the _features_compute_magnitude() function.

{a}, {b} and {r} features are are not enabled by default (and the - character is printed in the corresponding columns). To enable theses features you need to use the --cca-ell command line parameter. For more information about those features you can refer to the _features_compute_ellipse() function.

3.1.34.2. Table 3: List of associations between RoIs

# --------------------||---------------||------------------------||-----------
#         RoI ID      ||    Distance   ||  Error (or velocity)   ||   Motion
# --------------------||---------------||------------------------||-----------
# ----------|---------||--------|------||-------|-------|--------||-----------
#       t-1 |       t || pixels | rank ||    dx |    dy |      e || is moving
# ----------|---------||--------|------||-------|-------|--------||-----------
  {rid_t-1} | {rid_t} || {dist} |  {k} ||  {dx} |  {dy} |    {e} ||      {mov}

Each line corresponds to an association between one RoI at \(t - 1\) and at \(t\):

  • {rid_t-1}: id of the RoI in the table 1 (in the \(t - 1\) frame),

  • {rid_t} : id of the RoI in the table 2 (in the \(t\) frame),

  • {dist}: distance in pixels between the two RoIs,

  • {rank}: rank in the k-NN algorithm, if 1: it means that this is the closest RoI association, if 2: it means that this is the second closest RoI association, etc.,

  • {dx}: \(x\) distance between the estimated position (after motion estimation) and the real position (in frame \(t - 1\)),

  • {dy}: \(y\) distance between the estimated position (after motion estimation) and the real position (in frame \(t - 1\)),

  • {e}: euclidean distance between the estimated position and the real position,

  • {mov}: yes if the RoI is moving, no otherwise. The criteria to detect the motion of an RoI is: \(|e - \bar{e^1_t}| > \sigma^1_t\), with \(e\) the error of the current RoI, \(\bar{e^1_t}\) the mean error after the first motion estimation and \(\sigma^1_t\) the standard deviation after the first motion estimation.

If {mov} = yes then, {dx}, {dy} is the velocity vector and {e} is the velocity norm in pixel.

Note

{dx}, {dy}, {e} and {mov} are computed after the second motion estimation.

3.1.34.3. Table 4: Motion Estimation Statistics

# ------------------------------------------------------||------------------------------------------------------
#   First motion estimation (with all associated RoIs)  ||    Second motion estimation (exclude moving RoIs)
# ------------------------------------------------------||------------------------------------------------------
# ----------|----------|----------|----------|----------||----------|----------|----------|----------|----------
#     theta |       tx |       ty | mean err |  std dev ||    theta |       tx |       ty | mean err |  std dev
# ----------|----------|----------|----------|----------||----------|----------|----------|----------|----------
   {theta1} |    {tx1} |    {ty1} |{mean_er1}|{std_dev1}|| {theta2} |    {tx2} |    {ty2} |{mean_er2}|{std_dev2}

There is only one line in this table. It represents the motion estimation between frame \(t - 1\) and frame \(t\):

  • {theta}: the estimated rotation angle between frame \(t\) and frame \(t - 1\),

  • {tx} and {ty}: the estimated translation vector from frame \(t\) to frame \(t - 1\),

  • {mean_er}: the mean error of the associated RoIs,

  • {std_dev}: the standard deviation of the associated RoI errors.

The first estimation considers all the associated RoIs while the second estimation excludes the associated RoIs in movement. To be considered in movement, an RoI has to verify the following condition: \(|e - \bar{e^1_t}| > \sigma^1_t\), with \(e\) the error of the current RoI, \(\bar{e^1_t}\) the mean error after the first motion estimation and \(\sigma^1_t\) the standard deviation after the first motion estimation.

3.1.34.4. Table 5: List of Tracks

# -----------------||---------------------------||---------------------------||---------||-------------------
#       Track      ||           Begin           ||            End            ||  Object || Reason of changed
# -----------------||---------------------------||---------------------------||---------||    state (from
# -------|---------||---------|--------|--------||---------|--------|--------||---------||  meteor to noise
#     Id |  State  || Frame # |      x |      y || Frame # |      x |      y ||    Type ||    object only)
# -------|---------||---------|--------|--------||---------|--------|--------||---------||-------------------
   {tid} | {state} ||  {fbeg} | {xbeg} | {ybeg} ||  {fend} | {xend} | {yend} || {otype} ||          {reason}

Most of the columns of this table have been described in the Standard Output section, here we focus only on extra columns:

  • {state}: current state of the track (in a finite state machine): can be updated, lost or finished. First, when a track is created, its state is updated. Then, updated can become lost. lost can become either finished or updated. finished is a final state.

  • {reason}: reason of the classification from meteor to noise.

3.1.35. --vid-out-path

Type

STRING

Default

[empty]

Example

--vid-out-path ~/Videos/out_video.mp4

Output video path with the bounding boxes around the detected tracks (supports also a path to a sequence of images path/basename_%05d.jpg).

3.1.36. --vid-out-play

Type

BOOLEAN

Example

--vid-out-play

Open an SDL window to show the output video of the detected tracks (shows the bounding boxes around the detected tracks)

3.1.37. --vid-out-id

Type

BOOLEAN

Example

--vid-out-id

Work only if --vid-out-path or --vid-out-play is set. Display the track ids corresponding to the bounding boxes. Requires to link with OpenCV library (-DFMDT_OPENCV_LINK CMake option, see Section 2.2.1).