Research - User contributions [en]

Mean(csi)

2006-12-19T18:45:53Z

Bbogart:

== mean(csi) : A Time Series Analysis of CSI Season 1 Episodes ==

== Preamble ==

The original idea came from wanted to make an abstraction of an action movie. The pacing and intensity of movement would be the same but the image would not be photo-realalistic. Can an abstraction give the same sense of intensity as an action movie? I'm starting with CSI episodes are a smaller project to cut my teeth on.

== Hypothesis ==

There is a common style of pacing in CSI episodes. If this hypothesis is supported then an abstraction of that which is common between episodes will inform a time-based composition. If the null hypothesis is supported then the pacing from a single episode will be used to inform a time-based composition.

== Methods ==

* [[Initial Observations]]

* [[Early Results]]

* [[DataPlots]]

* [[Sonification]]

* [[Visualizations]]

* [[Software]]

== Conclusion ==

Both autocorrelation and crosscorrelation yeilded some qualitatively interesting data. There does seem to be some periodic element to the CSI episodes in themselves, but it is unclear if those elements are shared. With the data I have I am only able to support the null hypothesis that there is no periodic element that is shared between episodes.

== Future Work ==

On the qualitiative side there is lots of potencial for using the collecting pacing data to create interesting representations of the data. This will be the primary evolution of the project.

On the quantitiative side using ARIMA models and FFT should give a little more information of the nature of the pacing of the episodes and if there is any correlation accross episodes. At best I believe there is some grid of timing used in the editing of CSI episodes but these may actually be shared by other television programs. If I am able to find such a grid then I would compare that to other TV programs and a noise sample to make sure that grid is due to the CSI program, rather than than due to the DVD encoding process, or the video signal itself.

Mean(csi)

2006-12-19T18:36:21Z

Bbogart:

Software

2006-12-08T02:07:31Z

Bbogart:

Here is a list of the software I used to work on this project (so far).
The process has not been about writing a large peice of code but has been more about using the tools that are
available together rather than starting something from sratch. Each of the following pages gives details about how the tool was used and source-code.

* [[Python & PIL]]

* [[R]]

* [[Pure-Data & GEM]]

Pure-Data & GEM

2006-12-08T01:59:17Z

Bbogart:

Pure-Data is a graphical data-flow programming environment similar to Max/MSP. To find out more about Pure-Data visit the [[http://www.pure-data.info community page]]

Operations in Pure-Data are objects. Objects are selected by typing there name in a box. This is a lot like calling a function or instanciating a class. Where Pure-Data differs is that you relate these operations by drawing connections between them. A Patch is a collection of connections and operations. Here is a really simple example:

[[Image:pd_print.png]]

GEM (Graphics Environment for Multimedia) adds graphics and video functions to Pure-Data.

I used Pure-Data to generate the [[Visualizations]] and the [[Sonification]]

===The following patch converts the logfile created by python into a wav audio file:===

[[Image:data2wav.png]]

===This is the patch that was used to create the single-episode abstraction: ===

[[Image:single-episode.png]]

=== Patches can also be nested, so that an object can contain a patch. Here is main patch: ===

[[Image:multi-episode-parent.png]]

=== Where "episode-abstraction" contains the following patch: ===

[[Image:multi-episode-child.png]]

Pure-Data & GEM

2006-12-08T01:58:49Z

Bbogart:

Pure-Data is a graphical data-flow programming environment similar to Max/MSP. To find out more about Pure-Data visit the [[http://www.pure-data.info community page]]

Operations in Pure-Data are objects. Objects are selected by typing there name in a box. This is a lot like calling a function or instanciating a class. Where Pure-Data differs is that you relate these operations by drawing connections between them. A Patch is a collection of connections and operations. Here is a really simple example:

[[Image:pd_print.png]]

GEM (Graphics Environment for Multimedia) adds graphics and video functions to Pure-Data.

I used Pure-Data to generate the [[Visualizations]] and the [[Sonification]]

===The following patch converts the logfile created by python into a wav audio file:===

[[Image:data2wav.png]]

===This is the patch that was used to create the single-episode abstraction: ===

[[Image:single-episode.png]]

=== Patches can also be nested, so that an object can contain a patch. Here is main patch ===

[[Image:multi-episode-parent.png]]

=== Where "episode-abstraction" contains the following patch: ===

[[Image:multi-episode-child.png]]

File:Multi-episode-child.png

2006-12-08T01:58:19Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:58:05Z

Bbogart:

Pure-Data is a graphical data-flow programming environment similar to Max/MSP. To find out more about Pure-Data visit the [[http://www.pure-data.info community page]]

Operations in Pure-Data are objects. Objects are selected by typing there name in a box. This is a lot like calling a function or instanciating a class. Where Pure-Data differs is that you relate these operations by drawing connections between them. A Patch is a collection of connections and operations. Here is a really simple example:

[[Image:pd_print.png]]

GEM (Graphics Environment for Multimedia) adds graphics and video functions to Pure-Data.

I used Pure-Data to generate the [[Visualizations]] and the [[Sonification]]

===The following patch converts the logfile created by python into a wav audio file:===

[[Image:data2wav.png]]

===This is the patch that was used to create the single-episode abstraction: ===

[[Image:single-episode.png]]

=== Patches can also be nested, so that an object can contain a patch. Here is main patch ===

[[Image:multi-episode-parent.png]]

=== Where "episode-abstraction" contains the following patch:

[[Image:multi-episode-child.png]]

File:Multi-episode-parent.png

2006-12-08T01:56:54Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:56:18Z

Bbogart:

File:Single-episode.png

2006-12-08T01:53:05Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:52:54Z

Bbogart:

Pure-Data is a graphical data-flow programming environment similar to Max/MSP. To find out more about Pure-Data visit the [[http://www.pure-data.info community page]]

Operations in Pure-Data are objects. Objects are selected by typing there name in a box. This is a lot like calling a function or instanciating a class. Where Pure-Data differs is that you relate these operations by drawing connections between them. A Patch is a collection of connections and operations. Here is a really simple example:

[[Image:pd_print.png]]

GEM (Graphics Environment for Multimedia) adds graphics and video functions to Pure-Data.

I used Pure-Data to generate the [[Visualizations]] and the [[Sonification]]

The following patch converts the logfile created by python into a wav audio file:

[[Image:data2wav.png]]

This is the patch that was used to create the single-episode abstraction:

[[Image:single-episode.png]]

Pure-Data & GEM

2006-12-08T01:50:20Z

Bbogart:

File:Data2wav.png

2006-12-08T01:49:54Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:49:28Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:46:05Z

Bbogart:

Visualizations

2006-12-08T01:45:25Z

Bbogart:

I have made two time-based visual compositions for this project.

The first is a video that uses the image, sound and data from a particular episode to create an abstraction of that particular episode. This is d3-004, Episode 10:

* [http://www.sfu.ca/~bbogart/csi/video2.ogm episode-10-excerpt]

The second is a simplier visualization of the data alone (no sound or video from the episodes) of all 12 episodes:

* [http://www.sfu.ca/~bbogart/csi/video3.ogm all-episodes-excerpt]

Note that these files are in ogg/theora format. I suggest you play with with [[http://www.videolan.org vlc]]

Visualizations

2006-12-08T01:44:42Z

Bbogart:

I have made two time-based visual compositions for this project.

The first is a video that uses the image, sound and data from a particular episode to create an abstraction of that particular episode. This is d3-004, Episode 10:

* [http://www.sfu.ca/~bbogart/csi/video2.ogm episode-10-excerpt]

The second is a simplier visualization of the data alone (no sound or video from the episodes) of all 12 episodes:

* [http://www.sfu.ca/~bbogart/csi/video3.ogm all-episodes]

Note that these files are in ogg/theora format. I suggest you play with with [[http://www.videolan.org vlc]]

Visualizations

2006-12-08T01:39:22Z

Bbogart:

Mean(csi)

2006-12-08T01:34:27Z

Bbogart:

Mean(csi)

2006-12-08T01:33:36Z

Bbogart:

Mean(csi)

2006-12-08T01:33:27Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:32:30Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:32:17Z

Bbogart:

File:Pd print.png

2006-12-08T01:29:19Z

Bbogart:

Pure-Data & GEM

2006-12-08T01:28:45Z

Bbogart:

R

2006-12-08T01:21:24Z

Bbogart:

R is a statistical analysis and visualization package similar to the commercial "S".

I used R to load the data files created by the python program and create the plots.

== Command Summary ==

You can get help on R functions:

<pre>help.search("anova")</pre>

<pre>help(plot)</pre>

Load the time series Library into R:

<pre>library(tseries)</pre>

Load the python data file: ("d1-001" is the code I used for episodes, it refers to disk-1, title 1 of the DVD.

<pre>d1.001 = read.ts(file="data/d1-001.log")</pre>

Get some descriptives on the data:

<pre>summary(d1.001)</pre>

Do a line plot of the data in blue, with custom labels for the plot (main) and the Y Axix (ylab)

<pre>plot(d1.001,type="l",col="blue",ylab="Pixel mean of interframe difference",main="mean(csi) : Season 1, Epsiode 1")</pre>

Plot a nice histogram of the data with 3000 bins, for x values between 0 and 30:

<pre>hist(d1.001,col="blue",breaks=3000,xlim=c(0,30))</pre>

Find the location of one click of the mouse:

<pre>locator(1)</pre>

Save a 11 point plots vertically to a single PDF with 0.3 inch margins on all sides:

<pre>
pdf("Season-1-raw.pdf",width=8.5,height=11) # create a PDF file rather than plotting to screen.
par(mfrow=c(11,1),mai=c(0.3,0.3,0.3,0.3)) # set the page to 11 rows, 1 column, and 0.3 margins.
plot(d1.001,type="p",pch=".")
plot(d1.004,type="p",pch=".")
plot(d1.007,type="p",pch=".")
plot(d1.010,type="p",pch=".")
plot(d2.001,type="p",pch=".")
plot(d2.004,type="p",pch=".")
plot(d2.007,type="p",pch=".")
plot(d2.010,type="p",pch=".")
plot(d3.001,type="p",pch=".")
plot(d3.004,type="p",pch=".")
plot(d3.007,type="p",pch=".")
plot(d3.010,type="p",pch=".")
dev.off() # close the file
</pre>

Do a linear filter of the data, using a moving average function with equal weights on all coefficients, using blocks of 1min (3600 frames). This gives us the following coefficient: 1/(2*3600+1)=7201. "filter()" applies the function where the coeffecients are 1/7201 copied 7201 times. The repetion is done by the rep() function. The results of the filter are stored in the variable d1.004.filter_1min where any NA's (missing values) returned by the filter are ignored using na.exclude().

<pre>d1.004.filter_1min = na.exclude(filter(d1.004,filter=rep(1/7201,7201)))</pre>

We can write this filtered timeseries to disk:

<pre>write(d1.004.filter_1min,file="data/d1-004.filter_1min.data",ncolumns=1)</pre>

To autocorrelate (for all possible lag times) the filtered data from one episode to another you can use the "acf()" function. This makes a copy of the time series and time shifts it. It compares the correlation between the timeseries and the lagged version of itself. If you see a repeating pattern in the result then there us a periodic component for that particular lag.

<pre>acf(d1.004.filter_1min,lag.max=length(d1.004.filter_1min))</pre>

We can also do a crosscorelation, where two timeseries are compared to time lagged versions of each other rather than themselves. When we see a periodic pattern in this case it means that there is some temporal commonality between them. Usually this means one series contributes components to another.

<pre>ccf(d1.001.filter_1min,d1.004.filter_1min,lag.max=80000)</pre>

R

2006-12-08T01:09:21Z

Bbogart:

R

2006-12-08T01:05:04Z

Bbogart:

R is a statistical analysis and visualization package similar to the commercial "S".

I used R to load the data files created by the python program and create the plots.

== Command Summary ==

You can search the descriptions of R functions:

<pre>help.search("anova")</pre>

<pre>help(plot)</pre>

Load the time series Library into R:

<pre>library(tseries)</pre>

Load the python data file: ("d1-001" is the code I used for episodes, it refers to disk-1, title 1 of the DVD.

<pre>d1.001 = read.ts(file="data/d1-001.log")</pre>

Get some descriptives on the data:

<pre>summary(d1.001)</pre>

Do a line plot of the data in blue, with custom labels for the plot (main) and the Y Axix (ylab)

<pre>plot(d1.001,type="l",col="blue",ylab="Pixel mean of interframe difference",main="mean(csi) : Season 1, Epsiode 1")</pre>

Plot a nice histogram of the data with 3000 bins, for x values between 0 and 30:

<pre>hist(d1.001,col="blue",breaks=3000,xlim=c(0,30))</pre>

Find the location of one click of the mouse:

<pre>locator(1)</pre>

Save a 11 point plots vertically to a single PDF with 0.3 inch margins on all sides:

<pre>
pdf("Season-1-raw.pdf",width=8.5,height=11) # create a PDF file rather than plotting to screen.
par(mfrow=c(11,1),mai=c(0.3,0.3,0.3,0.3)) # set the page to 11 rows, 1 column, and 0.3 margins.
plot(d1.001,type="p",pch=".")
plot(d1.004,type="p",pch=".")
plot(d1.007,type="p",pch=".")
plot(d1.010,type="p",pch=".")
plot(d2.001,type="p",pch=".")
plot(d2.004,type="p",pch=".")
plot(d2.007,type="p",pch=".")
plot(d2.010,type="p",pch=".")
plot(d3.001,type="p",pch=".")
plot(d3.004,type="p",pch=".")
plot(d3.007,type="p",pch=".")
plot(d3.010,type="p",pch=".")
dev.off() # close the file
</pre>

Do a linear filter of the data, using a moving average function with equal weights on all coefficients, using blocks of 1min (3600 frames). This gives us the following coefficient: 1/(2*3600+1)=7201. "filter()" applies the function where the coeffecients are 1/7201 copied 7201 times. The repetion is done by the rep() function. The results of the filter are stored in the variable d1.004.filter_1min where any NA's (missing values) returned by the filter are ignored using na.exclude().

<pre>d1.004.filter_1min = na.exclude(filter(d1.004,filter=rep(1/7201,7201)))</pre>

We can write this filtered timeseries to disk:

<pre>write(d1.004.filter_1min,file="data/d1-004.filter_1min.data",ncolumns=1)</pre>

R

2006-12-08T01:02:21Z

Bbogart:

R is a statistical analysis and visualization package similar to the commercial "S".

I used R to load the data files created by the python program and create the plots.

Here is a summary of the commands I used:

Load the time series Library into R:

<pre>library(tseries)</pre>

Load the python data file: ("d1-001" is the code I used for episodes, it refers to disk-1, title 1 of the DVD.

<pre>d1.001 = read.ts(file="data/d1-001.log")</pre>

Get some descriptives on the data:

<pre>summary(d1.001)</pre>

Do a line plot of the data in blue, with custom labels for the plot (main) and the Y Axix (ylab)

<pre>plot(d1.001,type="l",col="blue",ylab="Pixel mean of interframe difference",main="mean(csi) : Season 1, Epsiode 1")</pre>

Plot a nice histogram of the data with 3000 bins, for x values between 0 and 30:

<pre>hist(d1.001,col="blue",breaks=3000,xlim=c(0,30))</pre>

Find the location of one click of the mouse:

<pre>locator(1)</pre>

Save a 11 point plots vertically to a single PDF with 0.3 inch margins on all sides:

<pre>
pdf("Season-1-raw.pdf",width=8.5,height=11) # create a PDF file rather than plotting to screen.
par(mfrow=c(11,1),mai=c(0.3,0.3,0.3,0.3)) # set the page to 11 rows, 1 column, and 0.3 margins.
plot(d1.001,type="p",pch=".")
plot(d1.004,type="p",pch=".")
plot(d1.007,type="p",pch=".")
plot(d1.010,type="p",pch=".")
plot(d2.001,type="p",pch=".")
plot(d2.004,type="p",pch=".")
plot(d2.007,type="p",pch=".")
plot(d2.010,type="p",pch=".")
plot(d3.001,type="p",pch=".")
plot(d3.004,type="p",pch=".")
plot(d3.007,type="p",pch=".")
plot(d3.010,type="p",pch=".")
dev.off() # close the file
</pre>

Do a linear filter of the data, using a moving average function with equal weights on all coefficients, using blocks of 1min (3600 frames). This gives us the following coefficient: 1/(2*3600+1)=7201. "filter()" applies the function where the coeffecients are 1/7201 copied 7201 times. The repetion is done by the rep() function. The results of the filter are stored in the variable d1.004.filter_1min where any NA's (missing values) returned by the filter are ignored using na.exclude().

<pre>d1.004.filter_1min = na.exclude(filter(d1.004,filter=rep(1/7201,7201)))</pre>

We can write this filtered timeseries to disk:

<pre>write(d1.004.filter_1min,file="data/d1-004.filter_1min.data",ncolumns=1)</pre>

R

2006-12-08T00:59:34Z

Bbogart:

R is a statistical analysis and visualization package similar to the commercial "S".

I used R to load the data files created by the python program and create the plots.

Here is a summary of the commands I used:

Load the time series Library into R:

<pre>library(tseries)</pre>

Load the python data file: ("d1-001" is the code I used for episodes, it refers to disk-1, title 1 of the DVD.

<pre>d1.001 = read.ts(file="data/d1-001.log")</pre>

Get some descriptives on the data:

<pre>summary(d1.001)</pre>

Do a line plot of the data in blue, with custom labels for the plot (main) and the Y Axix (ylab)

<pre>plot(d1.001,type="l",col="blue",ylab="Pixel mean of interframe difference",main="mean(csi) : Season 1, Epsiode 1")</pre>

Plot a nice histogram of the data with 3000 bins, for x values between 0 and 30:

<pre>hist(d1.001,col="blue",breaks=3000,xlim=c(0,30))</pre>

Find the location of one click of the mouse:

<pre>locator(1)</pre>

Save a 11 point plots vertically to a single PDF with 0.3 inch margins on all sides:

<pre>
pdf("Season-1-raw.pdf",width=8.5,height=11) # create a PDF file rather than plotting to screen.
par(mfrow=c(11,1),mai=c(0.3,0.3,0.3,0.3)) # set the page to 11 rows, 1 column, and 0.3 margins.
plot(d1.001,type="p",pch=".")
plot(d1.004,type="p",pch=".")
plot(d1.007,type="p",pch=".")
plot(d1.010,type="p",pch=".")
plot(d2.001,type="p",pch=".")
plot(d2.004,type="p",pch=".")
plot(d2.007,type="p",pch=".")
plot(d2.010,type="p",pch=".")
plot(d3.001,type="p",pch=".")
plot(d3.004,type="p",pch=".")
plot(d3.007,type="p",pch=".")
plot(d3.010,type="p",pch=".")
dev.off() # close the file
</pre>

Do a linear filter of the data, using a moving average function with equal weights on all coefficients, using blocks of 1min (3600 frames). This gives us the following coefficient: 1/(2*3600+1) "filter()" applies the function where the coeffecients are 1/7201 copied 7201 times. The repition is done by the rep() function. The results of the filter are stored in the variable d1.004.filter_1min where any NA's (missing values) returned by the filter are ignored using na.exclude().

<pre>d1.004.filter_1min = na.exclude(filter(d1.004,filter=rep(1/7201,7201)))</pre>

R

2006-12-08T00:51:26Z

Bbogart:

R

2006-12-08T00:46:59Z

Bbogart:

Software

2006-12-08T00:29:04Z

Bbogart:

Python & PIL

2006-12-06T21:22:27Z

Bbogart:

I had planned to do the computer vision aspect of this project with python-opencv (The python hooks into the Open Computer Vision library). I ended up abadoning this method because OpenCV does not have any simple tracking (center of gravity of moved pixels in the frame) only complex tracking. I looked at CAMShift and MHI algorithms. The CAMShift tracking the colour mean of an image, MHI is a motion history image that calculates vectors of motion in the image based on a gradient from old to new frames. Both these methods (and opencv in general) work best with static cameras, so these methods get total lost when there is a scene cut.

In order to get data more quickly after a month of work on opencv I decided to look at the Python Imaging Library (PIL) which gives you a nice number of functions for working with image files. This is not a real-time image processing library, though it is quite fast considering. PIL has not computer vision facilities built in. I ended up doing a very simple process where the program:

# Opens an image file.
# Opens the previous file in the sequence.
# Removes the colour data from the images
# Subtracts the two frames to get the difference
# Writes the mean value of the pixels in the image to a logfile.

I wrote two python programs for:

* Controlling ffmpeg to batch process the ripped VOB fiels into jpgs [http://www.sfu.ca/~bbogart/csi/vob2frame.py src]

* Image Processing [http://www.sfu.ca/~bbogart/csi/processFrames-PIL.py src]

Python & PIL

2006-12-06T21:19:28Z

Bbogart:

I had planned to do the computer vision aspect of this project with python-opencv (The python hooks into the Open Computer Vision library). I ended up abadoning this method because OpenCV does not have any simple tracking (center of gravity of moved pixels in the frame) only complex tracking. I looked at CAMShift and MHI algorithms. The CAMShift tracking the colour mean of an image, MHI is a motion history image that calculates vectors of motion in the image based on a gradient from old to new frames. Both these methods (and opencv in general) work best with static cameras, so these methods get total lost when there is a scene cut.

In order to get data more quickly after a month of work on opencv I decided to look at the Python Imaging Library (PIL) which gives you a nice number of functions for working with image files. This is not a real-time image processing library, though it is quite fast considering. PIL has not computer vision facilities built in. I ended up doing a very simple process where the program:

# Opens an image file.
# Opens the previous file in the sequence.
# Removes the colour data from the images
# Subtracts the two frames to get the difference
# Writes the mean value of the pixels in the image to a logfile.

I wrote two python programs for:

* Controlling ffmpeg to batch process the ripped VOB fiels into jpgs [[Media:Example.ogg]]

* Image Processing [[Media:Example.ogg]]

Python & PIL

2006-12-06T21:18:43Z

Bbogart:

I had planned to do the computer vision aspect of this project with python-opencv (The python hooks into the Open Computer Vision library). I ended up abadoning this method because OpenCV does not have any simple tracking (center of gravity of moved pixels in the frame) only complex tracking. I looked at CAMShift and MHI algorithms. The CAMShift tracking the colour mean of an image, MHI is a motion history image that calculates vectors of motion in the image based on a gradient from old to new frames. Both these methods (and opencv in general) work best with static cameras, so these methods get total lost when there is a scene cut.

In order to get data more quickly after a month of work on opencv I decided to look at the Python Imaging Library (PIL) which gives you a nice number of functions for working with image files. This is not a real-time image processing library, though it is quite fast considering. PIL has not computer vision facilities built in. I ended up doing a very simple process where the program:

# Opens an image file.
# Opens the previous file in the sequence.
# Removes the colour data from the images
# Subtracts the two frames to get the difference
# Writes the mean value of the pixels in the image to a logfile.

I solved two problems in python:

* Controlling ffmpeg to batch process the ripped VOB fiels into jpgs [[Media:Example.ogg] source]

* Image Processing [[Media:Example.ogg] source]

Software

2006-12-06T21:03:25Z

Bbogart:

* [[Python & PIL]]

* [[R]]

* [[Pure-Data & GEM]]

* [[dvdrip]]

Software

2006-12-06T21:03:13Z

Bbogart:

* Python & PIL

* R

* Pure-Data & GEM

* dvdrip

Mean(csi)

2006-12-06T21:01:53Z

Bbogart:

Sonification

2006-12-06T21:01:26Z

Bbogart:

Here are the sonifications of the CSI data.

This is a data-point to sample mapping with only normalization of the first season:

* [http://www.sfu.ca/~bbogart/csi/d1-001.wav Episode 1] [http://www.sfu.ca/~bbogart/csi/d1-004.wav Episode 2] [http://www.sfu.ca/~bbogart/csi/d1-007.wav Episode 3] [http://www.sfu.ca/~bbogart/csi/d1-010.wav Episode 4] [http://www.sfu.ca/~bbogart/csi/d2-001.wav Episode 5] [http://www.sfu.ca/~bbogart/csi/d2-004.wav Episode 6] [http://www.sfu.ca/~bbogart/csi/d2-007.wav Episode 7] [http://www.sfu.ca/~bbogart/csi/d2-010.wav Episode 8] [http://www.sfu.ca/~bbogart/csi/d3-001.wav Episode 9] [http://www.sfu.ca/~bbogart/csi/d3-004.wav Episode 10] [http://www.sfu.ca/~bbogart/csi/d3-007.wav Episode 11] [http://www.sfu.ca/~bbogart/csi/d3-010.wav Episode 12]

All episodes playing simultaneously in sync (scaled to so from 30 to 285Hz):

* [http://www.sfu.ca/~bbogart/csi/all-episodes.wav All Episodes]

Sonification

2006-12-06T19:55:02Z

Bbogart:

Sonification

2006-12-06T19:54:52Z

Bbogart:

Here are the sonifications of the CSI data.

This is a data-point to sample mapping with only normalization of the first season:

* [http://www.sfu.ca/~bbogart/csi/d1-001.wav Episode 1] [http://www.sfu.ca/~bbogart/csi/d1-004.wav Episode 2] [http://www.sfu.ca/~bbogart/csi/d1-007.wav Episode 3] [http://www.sfu.ca/~bbogart/csi/d1-010.wav Episode 4]
[http://www.sfu.ca/~bbogart/csi/d2-001.wav Episode 5] [http://www.sfu.ca/~bbogart/csi/d2-004.wav Episode 6] [http://www.sfu.ca/~bbogart/csi/d2-007.wav Episode 7] [http://www.sfu.ca/~bbogart/csi/d2-010.wav Episode 8] [http://www.sfu.ca/~bbogart/csi/d3-001.wav Episode 9] [http://www.sfu.ca/~bbogart/csi/d3-004.wav Episode 10] [http://www.sfu.ca/~bbogart/csi/d3-007.wav Episode 11] [http://www.sfu.ca/~bbogart/csi/d3-010.wav Episode 12]

Sonification

2006-12-06T19:54:35Z

Bbogart:

Here are the sonifications of the CSI data.

This is a data-point to sample mapping with only normalization of the first season:

* [http://www.sfu.ca/~bbogart/csi/d1-001.wav Episode 1] [http://www.sfu.ca/~bbogart/csi/d1-004.wav Episode 2]
[http://www.sfu.ca/~bbogart/csi/d1-007.wav Episode 3] [http://www.sfu.ca/~bbogart/csi/d1-010.wav Episode 4]
[http://www.sfu.ca/~bbogart/csi/d2-001.wav Episode 5] [http://www.sfu.ca/~bbogart/csi/d2-004.wav Episode 6]
[http://www.sfu.ca/~bbogart/csi/d2-007.wav Episode 7] [http://www.sfu.ca/~bbogart/csi/d2-010.wav Episode 8]
[http://www.sfu.ca/~bbogart/csi/d3-001.wav Episode 9] [http://www.sfu.ca/~bbogart/csi/d3-004.wav Episode 10]
[http://www.sfu.ca/~bbogart/csi/d3-007.wav Episode 11] [http://www.sfu.ca/~bbogart/csi/d3-010.wav Episode 12]

Sonification

2006-12-06T19:54:23Z

Bbogart:

Here are the sonifications of the CSI data.

This is a data-point to sample mapping with only normalization of the first season:

* [http://www.sfu.ca/~bbogart/csi/d1-001.wav Episode 1]
[http://www.sfu.ca/~bbogart/csi/d1-004.wav Episode 2]
[http://www.sfu.ca/~bbogart/csi/d1-007.wav Episode 3]
[http://www.sfu.ca/~bbogart/csi/d1-010.wav Episode 4]
[http://www.sfu.ca/~bbogart/csi/d2-001.wav Episode 5]
[http://www.sfu.ca/~bbogart/csi/d2-004.wav Episode 6]
[http://www.sfu.ca/~bbogart/csi/d2-007.wav Episode 7]
[http://www.sfu.ca/~bbogart/csi/d2-010.wav Episode 8]
[http://www.sfu.ca/~bbogart/csi/d3-001.wav Episode 9]
[http://www.sfu.ca/~bbogart/csi/d3-004.wav Episode 10]
[http://www.sfu.ca/~bbogart/csi/d3-007.wav Episode 11]
[http://www.sfu.ca/~bbogart/csi/d3-010.wav Episode 12]

Sonification

2006-12-06T19:53:24Z

Bbogart:

Here are the sonifications of the CSI data.

This is a data-point to sample mapping with only normalization of the first season:

* [http://www.sfu.ca/~bbogart/csi/d1-001.wav Episode 1]

* [http://www.sfu.ca/~bbogart/csi/d1-004.wav Episode 2]

* [http://www.sfu.ca/~bbogart/csi/d1-007.wav Episode 3]

* [http://www.sfu.ca/~bbogart/csi/d1-010.wav Episode 4]

* [http://www.sfu.ca/~bbogart/csi/d2-001.wav Episode 5]

* [http://www.sfu.ca/~bbogart/csi/d2-004.wav Episode 6]

* [http://www.sfu.ca/~bbogart/csi/d2-007.wav Episode 7]

* [http://www.sfu.ca/~bbogart/csi/d2-010.wav Episode 8]

* [http://www.sfu.ca/~bbogart/csi/d3-001.wav Episode 9]

* [http://www.sfu.ca/~bbogart/csi/d3-004.wav Episode 10]

* [http://www.sfu.ca/~bbogart/csi/d3-007.wav Episode 11]

* [http://www.sfu.ca/~bbogart/csi/d3-010.wav Episode 12]

DataPlots

2006-12-06T19:27:27Z

Bbogart:

* [[rawPlots]]

== High Resolution Plots ==

Season 1 Episode 1-12 (Top to Bottom):

* Scatter plot of the raw tracking data: [http://www.sfu.ca/~bbogart/csi/Season-1-raw.pdf]

* Moving average filtered (to 1min) tracking data: [http://www.sfu.ca/~bbogart/csi/Season-1-filtered_1min.pdf]

* Autocorrelation filtered data, lag=0 to 80000: [http://www.sfu.ca/~bbogart/csi/Season-1-acf.pdf]

* Crosscorrelation filtered data from episode 1-2, 2-3, 3-4 etc. [http://www.sfu.ca/~bbogart/csi/Season-1-ccf.pdf]

DataPlots

2006-12-06T19:27:07Z

Bbogart:

* [[rawPlots]]

== High Resolution Plots ==

Season 1 Episode 1-12 (Top to Bottom):

* Scatter plot of the raw tracking data: [http://www.sfu.ca/~bbogart/csi/Season-1-raw.pdf]

* Moving average filtered (to 1min) tracking data: [http://www.sfu.ca/~bbogart/csi/Season-1-filtered_1min.pdf]

* Autocorrelation filtered data, lag=0 to 80000: [http://www.sfu.ca/~bbogart/csi/Season-1-acf.pdf]

* Crosscorrelation filtered data from episode 1-2 2-3 3-4 etc. [http://www.sfu.ca/~bbogart/csi/Season-1-ccf.pdf]

DataPlots

2006-12-06T19:26:42Z

Bbogart:

* [[rawPlots]]

== High Resolution Plots ==

Season 1 Episode 1-12 (Top to Bottom):

* Scatter plot of the raw tracking data: [http://www.sfu.ca/~bbogart/csi/Season-1-raw.pdf]

* Moving average filtered (to 1min) tracking data: [http://www.sfu.ca/~bbogart/csi/Season-1-filtered_1min.pdf]

* Autocorrelation filtered data, lag=0-80000: [http://www.sfu.ca/~bbogart/csi/Season-1-acf.pdf]

* Crosscorrelation filtered data from episode 1-2 2-3 3-4 etc. [http://www.sfu.ca/~bbogart/csi/Season-1-ccf.pdf]

Early Results

2006-12-04T23:37:14Z

Bbogart:

* The following images are filtered using a moving average filter using the same weight for all coefficients. (1/7201 corresponding to 1 minute (3600 frames)):

[[Image:e1_filtered_1min.jpg]]

[[Image:e2_filtered_1min.jpg]]

* Here is the crosscorrelation between the these filtered versions:

[[Image:e1f_vs_e2f_crosscorrelation.jpg]]

So as great as these results look, turns out that the crosscorrelation between episode 1 and 2 is the greatest of all the episodes!

File:Pan vs object movement.jpg

2006-12-02T20:02:27Z

Bbogart: