G2-Update3
The Effect
Football player tracking with CG indicators
The Input
- Football game footage from side or goalpost view
- User's indication of the players
The Output
Original footage enhanced with CG indicators of tracked players
The Layers
- The raw football game footage
- CG indicators
The Pipeline
Progress
Research
Coding
- Used openCV to open an avi and iterate through the IplImage frames and playback in a openCV window
- Explored tracking demos that use the CamShift and MeanShift algorithms (Abandoned)
- Started writing code that uses cvGoodFeaturesToTrack and cvCalcOpticalFlow
- Optical flow for image sequences (since the field does not flow, this gives us additional information as to which areas contain players)
- Began development on effective tracking of multiple objects
- Significantly improved edge detection by filtering (might be used for feature matching – thanks Florian)
- Hough lines for every frame
- Homography from first frame to model
- Segmentation on following frame
- Intersections of Hough lines through Sobel edge detector
- Match intersections with model using estimation from previous homography
- Homography generation on frame sequence
- Segmentation on frame sequence
Images
First Frame
Model
Frame Warped by Homography
Overlay
Stadium Segmentation
Videos
Model with Matched Intersection Points and Field Lines
Segmented Field with Hough Lines and Intersection Points
Hough Lines (upgraded as seen above) and Optical Flow
Modified CamShift (Abandoned)
Timeline
04/10 – Project Update 3 (Automatic Homography and Segmentation Updates)
04/12 – FP_Preview Showing (Solidify Automatic Homography)
04/17 – Project Update 4 (Player Tracking)
04/26 – FP_Final Showing (CG and Clean up)
Links to this Page
- DVFX 2007 Groups last edited on 2 May 2007 at 12:51 am by coppola-win.cc.gatech.edu
- G2-Final Report last edited on 2 May 2007 at 5:18 am by coppola-win.cc.gatech.edu