BU CAS CS 585

Image and Video Computing

Fall 1996

Programming Assignment 4

Due before class on Tuesday, December 10

Due to limited time, this assignment has been modified. This is the new (reduced) version.

You are given digitized video sequences of cars moving on the Mass Pike. The video has been collected from a stationary camera. First develop a program that segments these moving cars from video sequences. Your program should produce a colored labeling of where each car is in each frame of the video.

Next, your program estimate the direction of motion for each extracted car. Show the direction as a displayed vector on the image. Also estimate the instaneous image velocity of the car. Direction and velocity estimates should be estimated using a robust method (like least squares or Kalman filtering). By this I mean, use all pixels belonging to a car in estimating the velocity. You are also encouraged to use multiple samples in time.

Mass Pike video data sets used for this project are available on the CGL cluster in the directories /scratch/sclaroff/seq1 and /scratch/sclaroff/seq2. More sequences may be gathered and posted next week.

Before you begin coding, first be sure to formally define the problem. What constraints can be used to simplify the problem and its solution? Include a detailed discussion of these in your write-up. Also give a detailed description of your approach and justify it in light of your problem formulation.

Test your approach on at least four different subsequences of 10 images. Report the results of these tests, including pictures. Discuss the performance of the particular technique employed, including strengths and weakness.

Note that the camera has introduced perspective foreshortening of the road in the image. How does this effect the accuracy of your estimates?

Lastly, could such a system address the needs of SmartRoute Systems as discussed in class? Give details of additional needed to bring the formulation from prototype into a working, dependable, system? If you feel that this method cannot meet the needs, you are encouraged to propose and defend an alternative method.

As before, the write-up is to be prepared in HTML format. Include the program source. Submit your complete HTML document using the submit program on CGL or on CSA.

Extra Credit

Use methods discussed in class and in the paper by Szeliski to undo the foreshortening effect produced by perspective projection of the road. Use the recovered parameters to warp the image sequence in such a way as to "undo" perspective. Estimate the motion parameters in the warped sequence. 

Page Created: Nov 22, 1996      
Last Modified: Nov 27, 1996
Maintained by: Stan Sclaroff