ImageDisplacements
ListVectorPlot
ListConvolve
Curl

Image Processing
Video Processing
Computer Vision

Detect Rotating Objects in a Video

Open in Cloud
Download Notebook

Apply vector field analysis on image displacements to capture video dynamics

Start with a video sequence of rotating wind turbines:

In[1]:=

v=Video

CloudObject[

]



Out[1]=

Use

ImageDisplacements

on the entire video sequence to obtain a vector field

that describes the optical flow that is essentially the 2D motion field of objects in the video:

In[2]:=

flow=VideoMapList[First@ImageDisplacements[#Image]&,v,2,1];

Compute the mean flow on all frames:

In[3]:=

meanFlow=Mean[flow];

Plot the vector field on an image of a video frame:

In[4]:=

ShowVideoExtractFrames[v,

frame

],ListVectorPlotTranspose@Reverse@meanFlow,



Out[4]=

Determine the amount of inherent rotation using the curl,

∇×

, of the vector field

In[5]:=

listCurl=flowListConvolve[{{0.5,0,-0.5}},flow〚All,All,2〛,{{1,2},{1,2}},"Fixed"]-ListConvolve[{{-0.5},{0.},{0.5}},flow〚All,All,1〛,{{1,2},{1,2}},"Fixed"];

Apply the result on the flow:

In[6]:=

curl=ImageAdjust@Image@(listCurl@meanFlow)

Out[6]=

Use

GaussianFilter

to obtain a blurred curl distribution. It provides a more robust result for large-scale rotations. Black stands for clockwise rotation and white for counterclockwise for positive rotation:

In[7]:=

blurredCurl=ImageAdjust[GaussianFilter[curl,8{3,1}]]

Out[7]=

Colorize the rotation map and superimpose it on top of the original image:

In[8]:=

ImageComposeVideoExtractFrames[v,

frame

],SetAlphaChannel[Colorize[blurredCurl,ColorFunctionColorData[{"TemperatureMap","Reverse"}]],ColorNegate[blurredCurl]]

Out[8]=

Related Symbols

ImageDisplacements
ListVectorPlot
ListConvolve
Curl

Publisher Information

Contributed by: Wolfram Staff