This is a pretty classic computer vision problem that combines detection, tracking, filtering, recognition, and logical parsing together into one system whose objective is to make sense of the comings and goings of people or objects within a scene. It is one of the few of the modules that requires some nominal level of system integration to get running properly. Matlab has Simulink code that does this for the case of abandoned object detection, which is documented online, so you can see one expected outcome of a surveillance system.

/* I also found this might be useful too: http://studentdavestutorials.weebly.com/particle-filter-with-matlab-code.html This website covers areas such as Bayes rule, Kalman filter and particle filter with short videos and Matlab implementation. The tracker parts(Kalman filter and particle filter) may be included to learning modules. */

1. Background estimation and subtraction.
2. Target tracking
3. Target modeling
4. Target recognition

The sequence below introduces one aspect of surveillance systems at a time. They direct you to Matlab code that sometimes implements multiple steps at a time. It is recommended that you implement each one individually to get a sense for what role it plays in the entire system, rather than just copy/paste the whole system.

Week #1: Setup, Data, and Basics
Explore the datasets available and select a couple of videos to use as the starting point for processing, testing, and debugging the surveillance system to be created as part of this learning module.

Meanwhile, check out this review of three simple background modeling methods. You are not advised to use the code provided when implementing the same in future activities. There are more efficient ways to do the same without resorting to as many for loops, plus the implementation needs to be packaged up for use by the overall system. Plus, the activities below utilize existing Matlab libraries to the extent possible or desirable.

As a first step, obtain the surveillance system class stub and also the main execution script. Modify the main loop code stub so it loads a video you've chosen, loops through the frames, displays the image associated the each frame, and quits when that's done. Naturally this code won't do any surveillance, but it will setup the system to do so.

As a second step, implement the basic background modeling detection step. Matlab has implementation of the mixtures of Gaussians adaptive background estimation algorithm. Just perform the estimation part and retrieve the binary foreground image. Modify the displayState function to display this output. When run, the system should display the source video, plus a binary image sequence associated to the detected objects.

Explore & Deliverables: How well is the background modeled? You can identify how well it works by examining the quality of the binary image sequence. Does it capture the target objects only? Are there more false positives or false negatives than you like? What did you do to get the best result possible (to what parameters)? You should turn in at least one image pair showing the input frame, plus the output frame after foreground detection (or with the mask as noted in the code stub).

Week #2: Foreground Object Extraction
With the background modelling step done, we have a means to identify regions of the image that do not conform to the expected scene. We will presume that these are all objects of interest to track, e.g. targets. Advance beyond the current foreground modelling step to include processing of the binary foreground image for extraction of detected targets and their bounding boxes. Extracting the foreground objects is really performing blob extraction from the binary foreground image.

Flesh out the overlayState function so that it can overlay the surveillance system output over the current image frame. In addition to keeping track of the total number of detected objects, like in the example, for each box plotted, plot the detection index associated to the box like here. You will also need to modify the functions that are invoked within the process member function. You may find that some will not be modified due to how the Matlab implementation works.

Apply the algorithm to the video S2.L1 Walking, View 001 from the PETS 2009 People Tracking Dataset. It may be a collection of images, in which case some modification of the main loop will be needed. Also apply to the couple of videos that you've selected (here, couple = exactly two). If one of them is the S2.L1 PETS 2009 video, then select another one to process.

Explore & Deliverables: Turn in state overlays of the processed video for the specified PETS 2009 dataset, plus for your two additional videos chosen. Discuss the performance of the foreground detection process, as well as the detection labelling for the targets over time.

• Are there any challenges that pop up regarding the box estimation process?
• How did you select your parameters?
• What are morphological operations?
• Explain the relationship between the erode, dilate, open, and close operations.

Naturally, you should be turning in your code in the submission document. The best is to provide code snippets of the class member functions that you modified as part of this activity. If you submit the entire class, then highlight the modified functions so that we can see what was done.

Week #3: Optimization-Based Data Association
Performing detection does provide a means to identify objects of interest versus the prevailing background image. However if we are interested in maintaining the identity of the objects, additional processing and logic is required. The simplest scheme simply considers the spatio-temporal history of the targets and tries to link the current detected objects to the previously detected objects. This form of data association is known as the assignment problem.

Two algorithms: Hungarian or Munkres' algorithm and Jonker-Volgenant Algorithm.

The net result should be similar to this Matlab demo on multiple object tracking, which also implements far more than what is described. You should strip the extras.

Week #4: Adding Temporal Dynamics via a Kalman Filter

We can do a better job handling things like occlusions, as well as improve the data association, by adding in a temporal filter. A simple, powerful method is to recursively estimate and correct the moving target dynamics through a Kalman filter. Matlab has a page on how to do so for object tracking along with a more general intro page to additional documentation.

• If you are interested in the details of the Kalman filter and their connection to code, this Matlab file exchange code tutorial may be of assistance.

Week #1: Differentiating People

Gaussian mixture model for targets using Expectation Maximization algorithm.

Week #2: Appearance-Based Data Association

Comparing models.

Week #3: Re-Identification

Using comparison to know when same person has re-entered.

TBD.

Sample videos from past teams:

• Youtube Channel for Team DT from ECE4580, Fall 2014.

Presentations by researchers in computer vision

• A good example of how to present your results. See the visuals that the Discrete-Continuous Optimization algorithm author uses. Plus this one.
• Another example.
• .. and another example.

Online talks

• Oral defense of Guang Shu @ UCF.

Advertisement Videos of companies that provide surveillance algorithms as a service:

• LightHaus Logic

• Other peoples tips on parameter selection for MOG foreground detector.
• Implement on a live feed? Does it run fast enough, or is that not necessary?

ECE4580 Learning Modules