dgt-2014

Title: Robust Computer Vision Techniques and its Application to Intelligent Transport Systems to Improve Road Safety, Mobility and Traffic Management.

Administrative reference: SPIP2014-1507

Funding entity: Funded by the Department of Traffic (Dirección General De Tráfico). Resolution of December 19(B.O.E. No. 309, 23 December 2014).

Introduction

Due to the progress of Computer Vision, video cameras have become a low-cost, low-maintenance sensor with a huge potential for provide rich information about traffic in comparison to other sensors. Consequently, video surveillance systems have acquired great relevance in the field of Intelligent Transport Systems, which aim to improve Road Safety, Mobility and Sustainability by means of suitable traffic monitoring and management.

Although during the last years the capabilities of computer vision techniques have considerably improved at every level, the complete automation of such video surveillance tasks, especially at high-level, still demands robust and efficient solutions for real operation environments.

The Directorate-General of Traffic (Dirección General de Tráfico, DGT) is the government department that is responsible for the Spanish transport network. DGT has a high number of video surveillance cameras deployed along the roads of the whole country, and the monitoring of all of these cameras by DGT personnel 24-hours 7-days is absolutely unfeasible. Consequently, the use of automatic surveillance systems, which are able to call the attention of human operators towards the events of interest, turns out to be a necessary innovation. Thus, the main goal of this project is to develop an automatic incident detection system specifically tailored to work with the already deployed video cameras (in particular, considering their video resolution and quality).

In this context, the overall result of the project is a prototype that can monitor the cameras currently deployed by the DGT and generate alerts when incidents occur. The most innovative feature of the prototype is that it allows the operators to move the cameras remotely and it keeps working with the new point of view without any type of calibration.

Summary of results

The necessary subsystems for the development of an automatic traffic incident detection have been designed and implemented by means of a high-level programming language; in particular:

    • Pre-processing (improving the quality of the received video signal).
    • Vehicle detection (movement) and road modeling.
    • Robust vehicle tracking and trajectory generation method.
    • Adaptation of the trajectories to the view of the camera (which can vary).
    • Incident detection.

The following videos illustrate some of these subsystems:

Figure 1. Adaptation of the trajectories

As mentioned, the greatest innovation introduced by the project is to take into account the possibility that operators can move cameras and leave them looking from a different point of view, so that the system must adapt to the new view, without any type of calibration. In this sense, the project contributions have been the following:

    1. Adaptations of the original detecting and tracking methods to run on low-quality and low-resolution video.
    2. Automatic detection of camera movement by the operator.
    3. Given a point of view of the camera, a novel technique has been developed to determine whether there already exists a trajectory model for that point of view.
    4. Geometric correction mechanisms to deal with similar point of views.

Mid and long-term potential results

Considering the possible impact of the results in the mid- and long-term, it is possible to think about new innovations that provide greater value and differentiation in the domain of intelligent transport systems; in particular:

    • Vehicle tracking system that provides instant speed and acceleration that can be used to generate statistical models for improving road safety. Also, when monitoring vehicles on the same lane, it would be possible to generate alerts in areas where the safety distance is low.
    • Detection of adverse weather conditions: early detection of heavy rain, hail or snow can draw the operator’s attention to road sections that are temporarily dangerous for these causes.
    • Trajectory models that provide information on the use of roads (lane use, drawn regular curves, etc.), so this would be used to improve their safety.
    • Detection of places and time slots where congestions usually happen and generation of information about their causes (how trajectories evolve to generate congestions).
    • Software to improve the quality of the video, which can be used beyond the scope of this project (for example, to send higher quality video to media).

Overall, the expected technological impact is great as long as the developed innovations provided with greater flexibility (no fixed cameras required) to current automatic incident detection systems based on computer vision.

Additionally, detection of incidents allows automatic recording of the corresponding video sequences for further study, which would open the door to other uses and applications of the captured images, which can range from analyzing incidents “a posteriori” to generation of models of normal vehicle behaviour in the different sections of the monitored roads, through the possibility of considering the available cameras as a connected network in order to extract useful information for the planning and management of road safety at global level.

 

 

Comments are closed