1. Sapienza Università di Roma Dipartimento di Informatica e Sistemistica A DISTRIBUTED VISION SYSTEM FOR BOAT TRAFFIC MONITORING IN THE VENICE GRAND CANAL D. Bloisi, L. Iocchi

2. 2 ARGOS Project Overview The ARGOS system is going to control a waterway of about 6 km length, 80 to 150 meters width, through 14 observation posts (Survey Cells). A utomatic R emote G rand C anal O bservation S ystem

3. 3 ARGOS Objectives management and evaluation of navigation rules traffic statistics and analysis security preservation of historical heritage (reduction of wave motion)

4. 4 ARGOS Functions optical detection and tracking of moving targets computing position, speed and heading of targets event detection (speed limits, access control, …) recording 24/7 video and track information (post- analysis) rectifying camera frames and stitching them into a composite view automatic PTZ tracking …

5. 5 Survey Cell 3 high resolution network cameras, a PTZ camera for zoom and tracking of the selected target, and 2 computers running the image processing and tracking software. The survey cells are installed on the top of several buildings leaning over the Grand Canal

6. 6 Survey Cells

7. 7 SC Software Architecture Background estimation Background subtraction Optical Flow Foreground Blobs Analysis Segmentation Center camera Right camera Segmentation List of observations Tracking Module Boat IDs Left camera

8. 8 Background Estimation Problems: - gradual illumination changes and sudden ones (clouds) - motion changes (camera oscillations) - high frequency noise (waves in our case) - changes in the background geometry (parked boats). Approach: - computation of color distribution of a set of frames - highest component form the background

9. 9 Background Estimation (2) Background Image computed from S (the image display only the higher gaussian values) Set S of 20 images from a camera Mask for cuttting off buildings from computation

10. 10 Background Subtraction current frame background image foreground image THRESHOLD (based on illumination conditions) blobs (Binay Large OBjectS) >

11. 11 Optical Flow Computation We use a sparse iterative version of Lucas-Kanade optical flow in pyramids ([Bouget00]). It calculates coordinates of the feature points on the current video frame given their coordinates on the previous frame. The function finds the coordinates with sub-pixel accuracy. Every feature point is classiefied into one of the four principal directions NE, NW, SE, SW. [Bouguet00] Jean-Yves Bouguet. Pyramidal Implementation of the Lucas Kanade Feature Tracker. previous framecurrent frame optical flow image (a particular) NW direction

12. 12 Segmentation Exploiting the foreground image and the optical flow image, for every blob we obtain  its centroid (that is (x, y) position into the current frame)  its direction (and consequentely the probability of under segmentation if the blob is classified into more than one of the principal directions)  its ellipse approximation (and consequentely its dimensions in meters through homography matrices) Blob filtering: If a blob is too small according to the minimal dimension a boat must be in order to navigate the Gran Canal) Under segmentation: If a blob has two or more directions we compute the center of mass and the variance for every of the four predetermined principal direction.

13. 13 Segmentation (2) blue → NW direction red → NE direction green → SE direction centroid ellipse center of mass

14. 14 Rek-means

15. 15 Rek-means (2)

16. 16 Tracking module Single-hypothesis Tracking We use a set of Kalman Filters (one for each tracked boat). Data Association: Nearest Neighbor rule Track formation: unassociated observations Track deletion: high covariance in the filter Multi-hypothesis Tracking Track splitting: in ambiguous cases (data association has multiple solutions) Track merging: high correlation between tracks

17. 17 Multi hypothesis tracking (2) 3 tracks (240, 247, 285) only 1 actual observation (285) 240 285 247

18. 18 Rectification

19. 19 Unified Views

20. 20 Panoramic view PTZ Camera

21. 21 Example

22. 22 DENSITA’ DI TRAFFICO – TEMPO REALE 1 2 4 3 8 5 7 6 10 9 12 11 13 DENSITA' MEDIE E MASSIME DEL TRAFFICO 02/11/2006 ore 11,30 Numero Totale Imbarcazioni in Canal Grande: 121 TrattoDaADensità media Densità max 1Ponte LibertàScomensera 5 8 2 Ponte Calatrava 6 12 3Ponte CalatravaFerrovia 8 10 4FerroviaCannaregio 10 18 5CannaregioSanta Fosca 6 18 6Santa FoscaCa D'oro 4 4 7 Rialto 12 16 8RialtoSan Silvestro 58 9S.SilvestroSan Tomà 14 26 10San TomàCa' Rezzonico 21 25 11Ca' RezzonicoAccademia 8 9 12AccademiaSalute 14 18 13SaluteBacino S.Marco 8 8

23. 23 Example

24. 24 Experimental Evaluation on-line, evaluation is performed during the actual operation of the system; recorded on-line evaluation is performed on a video recording the output of the system running on-line; off-line evaluation is performed on the system running off- line on recorded input videos.

25. 25 Online Evaluation FN: False negatives, i.e. boats not tracked FP-R: False positives due to reflections (wrong track with a random direction) FP-W: False positives due to wakes (wrong track following the correct one)

26. 26 Counting Evaluation COUNTING EVALUATION TEST A virtual line has been put across the Canal in the field of view of a survey cell, the number of boats passing this line has been counted automatically by the system nSys, and the same value is manually calculated by visually inspection n, the average percentage error is then computed as ε = | nSys – n | / n An additional error measure is calculated by considering the probability of making an error in counting a single boat passing the line where δ (·) is 0 when the argument is 0 and 1 otherwise.

27. 27 Counting Evaluation (2)

28. 28 Speed and velocity tests