Objectives: The goal of this project is to learn the basic concepts and techniques related to video sequences processing, mainly for surveillance applications. We will focus on video sequences from outdoor scenarios, with the application of traffic monitoring in mind. The main techniques of video processing will be applied in the context of video surveillance: moving object segmentation, motion estimation and compensation and video object tracking are basic components of many video processing systems. In a first stage, moving object segmentation will be tackled considering scenarios with static camera. Afterwards, camera motion will be considered. Tracking of the moving objects can be performed in both scenarios. The tracking result provides high level information that can be analysed for traffic monitoring. The learning objectives for the students are the use of pixel based statistical models (such as mixture of gaussians) for modeling a scene background and for moving object segmentation, the development of optical flow estimation methods for camera motion compensation, and techniques for object tracking (ranging from simple blob analysis to more complex techniques based on filtering and probabilistic data association). The performance of the developed techniques will be measured using standard metrics for video analysis.
This work is composed by 5 sessions, which one have its goals in order to step bu step perform the final Road traffic monitoring:
- Session 1: Assessment of Foreground Extraction and Optical Flow.
- Session 2: Background estimation.
- Session 3: Foreground segmentation.
- Session 4: Video Stabilization.
- Session 5: Region tracking.
- Slides of the project Google, [PDF](Final Slides.pdf)
- Project Report
In the following video we can see the different result that we obtain in this final part of the project. With our video we also include the mask without some of the methods to see the influence of these ones.
Video of Traffic:
We can see how don't work properly and the tracking system can't mantain properly the tracking of the vehicles, so need to be adjusted because we use the main system like the kalman filter and doesn't seem to be compatible entirely in order to mantain ID and tracks properly.
Video of Highway:
Position of our video: Av. de Serragalliners, 112, 08193 Cerdanyola del Vallès, Barcelona
<iframe src="https://www.google.com/maps/embed?pb=!1m16!1m12!1m3!1d1056.4726676988516!2d2.1129707569473566!3d41.5008951892322!2m3!1f0!2f0!3f0!3m2!1i1024!2i768!4f13.1!2m1!1sgoogle+maps!5e0!3m2!1sen!2ses!4v1486580514671" width="600" height="450" frameborder="0" style="border:0" allowfullscreen></iframe>As we can see on our video, in real case all the techniques that we have learn are necessary to improve the result. Because the influence of the daylight we can have problem of shadows or because the wind the camera could jig so these methods show they utility when comparing different situation of the open world.