Skip to content

mcv-m6-video/mcv-m6-2019-team4

Repository files navigation

Road traffic monitoring

Module 6 Project from Master in Computer Vision Barcelona program.

The goal of this project is to learn the basic concepts and techniques related to video sequences mainly for surveillance applications. More particularly, we develop techniques to track vehicles across different cameras which could be used for traffic monitoring (estimate speed, traffic density, etc.).

Scope

  • Use of statistical models to estimate the background information of the video sequence
  • Use of deep learning techniques to detect the foreground
  • Use optical flow estimations and compensations
  • Track detections
  • Analyze system performance evaluation

Applicability

Any problem where video sequence analysis is applied to automatically track objects.

How to run

Package dependencies are managed using pipenv. Check its documentation or its repository.

For every week there are a set of tasks to do. They can be found organized in package and modules for weeks and tasks respectively, i.e. week1\task1.py. All of them can be run from main.py which is used to collect all work done.

Main tasks done per week

  • Week1: analysis of the AICity challenge 2019 dataset. Creation of Optical Flow metrics (Mean Squared Error in Non-occluded pixels (MSEN); Percentage of Erroneous Pixels which are Not occluded (PEPN)).
  • Week2: background subtraction/foreground detection with single gaussian models: non-adaptive vs adaptive background model.
  • Week3: Object Detection (off-the-shelf + fine-tuning). Introduction to Object Tracking (Multi-track single camera).
  • Week4: Optical Flow estimation (off-the-shelf + block matching-based one). Video stabilization via Optical Flow vs State-of-the-art.
  • Week5: improve tracking with Optical Flow. Start thinking how to combine tracks (merge/split/swap IDs) among different cameras (Multi-track Multiple Camera).
  • Week6: test different multicamera setups and assess MTSC and MTMC against the state-of-the-art presented in the challenge's paper.

Results

The final presentation summarizing the techniques employed for vehicle tracking and re-identification can be accessed here.

The report in a paper-like fashion can be accessed here.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •