Technologies of Vision

SCOCA at work

SCOCA is video-based junction monitoring system that is able to extract and collect traffic data in real-time by analysing video sequences acquired from monocular pole-mounted cameras.

The applicative goal of the system is the statistical data collection about traffic, such as the number of vehicles of each class that cross the junction, their average speed, the intensity and the direction of the traffic flow with respect to the time, the origin-destination distribution map.

* First step: Initialize the intersections to be monitored (off-line operation): the operator inputs some camera parameters, specifies the portion of the image to be monitored and selects, along its boundary, a set of points which define the lanes of interest for the turning movement statistics (see Figure below - left).
Second step: the operator can schedule the data collection sessions from the various cameras. For each view the traffic manager inputs the starting time and the duration of the sessions (see Figure below - right). A module controls the compatibility of the selected time intervals by managing an agenda of events.

init of one intersection panel for booking the 
analysis sessions
Left: panel to input the road section to be monitored and the in/out virtual gates of interest.
Right: panel to book the starting time and duration of the analyses on the different junctions.

* An example: 5 minutes video labelled with the SCOCA output:

Here, pedestrian groups are the major cause of error. See the output OD map with the number of vehicles for each enter/exit gates.

* The data extractor analyses the images acquired from the camera (mpeg flow, sequence of jpeg, or other formats). A module maintains an updated estimation of the background image. Edges of the difference between current image and background are clustered to detect new entering vehicles and track them through the scene. Tracking is performed following a feature tracking paradigm in subsequent frames and by testing the overlap of regions in the predicted position every few frames (see figure below). The classification of the passed vehicles is a mixture of model-based and feature based modes.

detection localization ctracking
SCOCA data extactor work. Left: detection of new entering object and region-based tracking (every 5-7 frames). Center: green squares indicate the choice of particular features of interest. Right: in the subsequent frame, such features are searched in a neighborhood (red square) of the expected position (green square).

* As a result, the system counts the vehicles, estimates their speed, classify them into 7 classes (bicycle, motorcycle, car, van, lorry, urban bus, extraurban bus), and populates the global and class enter/exit map. To distinguish bicycles from motorcycles is a peculiar feature of this system. Isolated pedestrian are also captured.

SCOCA output. A panel provides a syntetic view of the analysis result. Number and average speed for each class of vehicles, local Origin/Destination flow map, histograms of vehicles passing from each virtual gate distighuished by class. For each vehicle, rough data is also stored for subsequent analysis and statistical data characterization.

* Data stored by SCOCA can be aggregate and analysed to find interesting traffic statistics. We have implemented a package, called Statistic Suite, that provides: traffic volume and average speed through the time, for each class and for each lane, distance between vehicles, occupancy for each specified zone and class, anomaluos paths analysis. A powerful interface allows the user to investigate the traffic phenomenon through various customizable graphs and values computation.

* SCOCA inside functioning is briefly described in the SCOCA research page and in the publications.

* The main publication describing the whole system is: S. Messelodi, C.M. Modena and M. Zanin, A computer vision system for the detection and classification of vehicles at urban road intersections, Pattern Analysis and Applications, 8(1-2):17-31, September 2005 [abstract] [TechRep pdf]

* Traffic WebCams in Trento (external link)

Are you interested in this technology?
Please contact Stefano Messelodi (messelod (at) fbk . eu)

Page Maintainer: Carla Maria Modena [modena (at) fbk . eu]
  tev   |   research   |   technologies   |   people   |   links   |   publications   |   students   |   private