Intelligent Surveilance Systems
DOWNLOAD THE FINAL PAPER HERE
Supervisor:
- Michael Hofmann - mhofmann @ science
Members:
- Paul Koppen - wpkoppen @ science
- Nicholas Piël - npiel @ science
- Štefan Konečný - konecny @ centrum . sk
Links
Plan
- Read relevant material for Chamfer system [5], hierarchical object detection [9], assembly techniques [1, 3, 8], and survey material [6].
- Download and install OpenCV library [2]. Get familiar with the programming with the library [10].
- Practice usage of part detectors. Run simple tests on several images. Make a module that can be later integrated with the aggregated algorithm
- Develop (very simple) labeling application. The application should load a series of images (e.g. all *.ppm files from a given folder), present a GUI that allows a user to draw rectangles, and save parameters of the rectangles. For each image: number and coordinates of all rectangles in a text of Matlab file.
- Label around 200 images (2 minutes/image x 200 images = 7 hours = 3 days x 2.5 hours/day). It is recommended that this step is performed quite early in the pro ject, since the evaluation of detectors requires labeled images.
- Given labeled images, test individual detectors in a systematic setup (ROC curves)
- Design/implement/test assembly algorithm (Several versions are welcome).
- Start simple. Use the evaluation results of part detectors to decide which parts are most reliable.
- Think of constraints on distance and size of one part relative to the other.
- Think of color (appearance) similarities between body parts.
- See [1, 8, 7] for inspiration.
- Perform systematic tests on labeled images. Compare results with individual detectors. Compare processing time.
- Write paper, prepare a presentation.
Papers
You can download a subset of the reading list as a zip file.
Selection of papers:
- [1] Agarwal - Learning to detect objects in images via a sparse, part based representation.
- [3] Felzenszwalb - Structures for object recognition
- [4] Gavrilla - Visual Analysis of human movement
- [5] Micilotta - Detection and tracking of humans by probalistic body part assembly
- [6] Mikolajczyk - Human detection based on probalistic assembly of robust part detectors
- [7] Mohan - Example-based object detection in images by components
- [8] Mori - Recovering human body configurations: Combining segmentation and recognition
- [9] DOAS 2007: Detecting humans by combining human part-detectors in an urban setting
- [10] Viola - Robust Real Time Face detection (!)
- [11] Wu - Detection and tracking of multiple partially occluded humons by bayesian combination
Schedule
- week 2: kick-off meeting (Monday 7th, 11:00-17:00), Kruislaan 403, room F0.09.
- week 2: first group meeting (Thursday 10th 15:00-17:00), Kruislaan F0.04
- week 3: progress meeting (Tuesday 15th, 13:00-16:00), Kruislaan 403, room F0.13
- week 5: deadline draft article (Monday 28th, 9:00, pdf on website)
- week 5: deadline review article (Tuesday 29th, 12:00, see form)
- week 5: deadline final article (Thursday 31th, 16:00, pdf on website)
- week 5: mini-conference (Friday 1st, 10:00-14:00), Kruislaan 403, room F0.13.
changed December 4, 2009