1 Video Surveillance systems for Traffic Monitoring Simeon Indupalli

2 Presentation Overview Video surveillance systems. Traffic monitoring issues. Object tracking techniques. Vehicle tracking strategies. A real time system Explanation. Future Work

3 what is video surveillance? Present Implementations?  Human detection systems.  vehicle monitoring systems. Advantages of video surveillance?  Keep track of information video data for future use.  Helpful in identifying people in the crime scenes etc.. Disadvantages of the present system?  It’s difficult to maintain heavy amount of raw video data  Human interaction.  Require higher bandwidth for transmitting the visual data.

4 Video surveillance in the context of Computer Vision Detection and tracking of moving objects are the important tasks of the computer vision. The video surveillance systems not only need to track the moving objects but also interpret their patterns of behaviours. This means solving the information and integration the pattern. Advantages  Minimizes the user interaction.  Less amount of prohibitive bandwidth.  Minimizes the cost and time.

5 Need for Traffic Monitoring To reduce the traffic congestion on highways Reduce the road accidents Identifying suspicious vehicles. Etc..,

6 Traffic Monitoring in Computer Vision The quest for better traffic information, an increasing reliance on traffic surveillance has resulted in a better vehicle detection. Taking some intelligent actions based on the conditions. Traffic scene analysis in 3 categories.  A strait forward vehicle detection and counting system.  Congestion monitoring and traffic scene analysis.  Vehicle classification and tracking systems which involve much more detailed scene traffic analysis.

7 Responsibilities of reliable Traffic Monitoring System Adaptive to changes in the real world environments Easy to set up Capable of operating independently of human operators. Capable of intelligent decisions. Capable of monitoring multiple cameras and continuous operation. Reasons for unsuccessful implementation**

8 A Traffic Monitoring System

9 Object Classification Shape based classification.  Image blob area, blob bounding box  Classification based on above info. Motion-based classification.  Human motion shows periodic property.  Time frequency analysis applied.  Residual flow taken under consideration.

10 Object tracking strategies (I)* Background subtraction  Difference between the current image and the reference background image in a pixel by pixel fashion.  Sensitive to the background changes  Wallflower principles for effective background maintenance.

11 Object tracking strategies (II) Temporal differencing  Moving objects changes intensity faster than static ones  Uses consecutive frames to identify the difference.  Adaptive to dynamic scene changes  Problems in extracting all relevant features.  Improved versions uses three frames instead of two

12 Object tracking strategies (III) Optical flow  To identify characteristics of flow vectors of moving objects over time.  It’s used to detect independently moving objects in presence of camera.  Requires a specialized hardware to implement. Optical flow of moving objects Meyer et al

13 Vehicle detection techniques Model based detection Region based detection Active contour based detection Feature based detection

14 Vehicle detection technique (I) Model based Tracking  The emphasis is on recovering trajectories and models with high accuracy for a small number of vehicles.  The most serious weakness of this approach is the reliance on detailed geometric object models. Disadvantage  It is unrealistic to expect detailed models for all vehicles that could be found on the roadway

15 Vehicle detection technique (II) Region based tracking  It detects each vehicle blob using a cross correlation function.  Vehicle detection based on back ground subtraction. Disadvantage  Difficult to detect the vehicles under congested traffic, because vehicles partly occlude with one another Potential segmentation problem

16 Vehicle detection technique (III) Active contour based detection  Tracking is based on active contour models, or snakes.  Representing object in bounding contour and keep updating it dynamically.  It reduced computational complexity compared to the region based detection. Disadvantage:  The inability to segment vehicles that are partially occluded remains a problem. Bounding counters

17 Vehicle detection technique (IV) Feature based detection  Tracks sub-features such as distinguishable points or lines on the object  Effectiveness improved by the addition of common motion constraint. Features are grouped together based on common motion, avoiding segmentation problem due to occlusion

18 A typical vehicle tracking procedure

19 Wallflower Principles & Practice of Background Maintenance. Moved objects Time of day Light switch Waving trees camouflage Foreground capture Stopped car Moving car Shadows Bootstrapping

20 Wallflower: Three levels of abstraction  Pixel level Maintains models of back ground of each individual pixel. Processing makes the preliminary classification between foreground and background Dynamic to scene changes.  Region level Emphasis is on interrelationship between the pixels Helps to refine raw classification at pixel level  Frame level It watches for the sudden changes in the large parts of the image and swaps in alternative background models.

21 A real time traffic monitoring system Feature based tracking algorithm Camera calibration Feature detection Vehicle tracking Feature grouping Benjamin Coifman, Jitendra Malik, David Beymer

22 Offline camera definition Line correspondences for a projective mapping. A detection region near the image bottom and an exit region at the image top And multiple fiducial points for camera calibration Based on the above information the system computes the homography between the image coordinates(x,y) and the world coordinates(X,Y)

23 On-line tracking and grouping Detector  Detecting corners at the bottom of image, where brightness varies in more than one direction.  Detection operationalzed by the points in the image I Tracker  Uses kalman filters to predict the velocity in the next image.  Normalized correlation is used to search the small region of image. Group  Grouper uses common motion constraint.  Once all the corner features are identified they are grouped together.  Monitoring the distance between the point d(t)=P1(t)-p2(t)

24 Sample corner features identified by the tracker Sample feature tracks from the tracker Sample feature groups from the tracker 1 2 3

25 Conclusion & Future Work The real time traffic surveillance system is still under research due to the background maintenance problem and occlusion. Better Background maintenance Solving occlusion problem

26 References: A Survey on visual surveillance of object motion and behaviour – HU et al Transportation research part-c/ A real time computer vision system for Traffic monitoring and vehicle tracking – B.coifman, J.Malik etc.. Steps towards cognitive vision system – H.Nagel, IAKS Karlsruhe. VSAM project – Carneigh Mellon University Wallflower Principles and practices – Microsoft Research group.