Showcase: Computer Vision

2014

Efficiently obtaining a reliable coronary artery centerline from computed tomography angiography data is relevant in clinical practice. 

This project aims to implement centerline extraction semi-automatic methods. It is based on two seed points (start and end) placed by the radiologist. Then, It extracts interactively the centerline using a medialness image [1][2] and a minimum cost path algorithm [3]. All code has been developed entirely in C++, using ITK and VTK libraries and the medical imaging framework CreaTools [4]. Currently, further medialness techniques are being implemented in order to do a comparison between them. 

References

[1] Esteban Correa-Agudelo, Leonardo Flórez-Valencia, Maciej Orkisz, Claire Mouton, Eduardo E. Dávila Serrano, Marcela Hernández Hoyos, "A Modular Workflow Architecture for Coronary Centerline Extraction in Computed Tomography Angiography Data", ICCVG 2014, Warsaw, Poland. (doi:10.1007/978-3-319-11331-9_19).

[2] CreaTools http://www.creatis.insa-lyon.fr/site/en/CreaTools_home

Further information:

http://www.creatis.insa-lyon.fr/

Master Thesis: Chapter 3

2014

This project aims at creating a computer vision-based framework for traffic analysis. Vehicle counting and tracking are based on MoG segmentation strategy [1] and overlapping heuristic of foreground element bounding box. The classification module uses the BoW+SVM approach [2] to successfully separate vehicle classes. Statistics are available in offline/online scenarios using a publisher-subscriber API (zeroMQ). Current work is mainly focused on real-time optimization (currently running at 33-34fps@fanless microITX) and ANPR module detection. 

References

[1] Z.Zivkovic, F. van der Heijden, Efficient Adaptive Density Estimapion per Image Pixel for the Task of Background Subtraction, Pattern Recognition Letters, vol. 27, no. 7, pages 773-780, 2006. The algorithm similar to the standard Stauffer&Grimson algorithm with additional selection of the number of the Gaussian components based on: Z.Zivkovic, F.van der Heijden, Recursive unsupervised learning of finite mixture models, IEEE Trans. on Pattern Analysis and Machine Intelligence, vol.26, no.5, pages 651-656, 2004.

[2] Visual Categorization with Bags of Keypoints by Gabriella Csurka, Christopher R. Dance, Lixin Fan, Jutta Willamowski, Cedric Bray, 2004.

[3] http://sirius.utp.edu.co

2012

The task of pedestrian tracking is a key component of video surveillance and monitoring systems. In this video, I play with the meanshift algorithm for tracking a pedestrian by combining with Kalman[2]. Firstly, transform the current frame from RGB space to the CIE1976 (L*a*b*) color space to distinguish the target from the background and other pedestrians. Secondly, use the adaptive meanshift [3] algorithm to calculate the window size and the position of the person. Finally, I use the Kalman filter to avoid the Camshift failure in some occlusion circumstances. Experiments on various video sequences illustrate the proposed algorithm performs better than using only the Camshift approach. 

References

[1] D. Comaniciu and P. Meer, “Robust Analysis of Feature Spaces: Color Image Segmentation,” CVPR’97, pp. 750-755.

[2] Gary R. Bradski, "Computer Vision Face Tracking For Use in a Perceptual User Interface", Intel Technology Journal, No. Q2. (1998).

[3] R. E. Kalman. "A New Approach to Linear Filtering and Prediction Problems", Transactions of the ASME – Journal of Basic Engineering, No. 82 (Series D). (1960), pp. 35-45.