Facebook  Twitter  Linkedin  YouTube
Sunday , 24 March 2024

MediaTunnel A real-time Traffic Incident Detection System

System Architecture

The system consists of CCTV cameras monitoring the tunnel, a set of analysers, a server and workstations. All system components are connected together using a local Ethernet network that uses the TCP/IP protocol. Each analyser, which is a high-grade industrial computer, consists of a computation unit, the image processing software, a video signal digitisation system and a video storage unit connected to a local network. Each analyser acquires, decodes and samples the IP video signals from the CCTV cameras, and acquires and digitises analogue video signals as well. It also records video sequences captured using the cameras on a periodic basis, a periodicity which can be set by the user. The server stores all the incident alarms, incident video clips and other relevant traffic data in a centralised database. It is connected to workstations and in some cases, to a SCADA (Supervisory Control and Data Acquisition) system in a Traffic Management Centre (TMC). Each workstation has the MediaTunnel monitoring software installed on it which can be periodically updated through the server, thus maintaining it up-to-date. The workstation displays incident alarms and other relevant traffic data.

We needed an Incident Detection System with good capabilities with IP streams and MediaTunnel is good at this. The redundancy features of the system added another level of safety. Plus, the ability of MediaTunnel to say if the incident has occurred in fluid or congested traffic is important for traffic operators. We also found the number of traffic operators needed to monitor the traffic inside the tunnel has decreased, and most of the incident alarms given off by MediaTunnel have been reliable. — Rubens Rodrigues

At system start up, the system algorithm builds up a reference image of the background of the monitored area, the image being continuously updated as the system receives new information. The algorithm works by extracting information of objects in the various video images captured by CCTV cameras. The area monitored by the CCTV cameras can be easily defined by the operator. Objects within the images could be vehicles and pedestrians, or simply artefacts such as shadows which could generate false alarms by the system. The algorithm avoids false alarms by using a double comparison method to detect vehicles within each image. The use of morphological filters that assign a distinct marker to each object in motion enables the system to separate vehicles from other objects in motion such as pedestrians. The algorithm also uses trajectory analysis of the objects in motion to rule out potential events that do not correspond to true traffic incidents. This is done by tracking the object by analysing the movement of its marker throughout the sequence of images to build up a time and space trajectory of the object. During periods of traffic congestion, vehicles are often occluded, and the cameras find difficulty in tracking them. In these situations an active occlusion management mechanism preserves the tracking markers, and the system resumes normal tracking when visibility is restored. This tracking and system analysis forms the basis of the traffic incident detection capabilities of the system.

The analyser stores in its buffer recorded digital video clips of a certain time duration preceding the incident. This duration can be adjusted. When an incident occurs, the video clip is extended either until the end of the incident or for a given time after the incident. The operator can then immediately display the video clips of the time periods before and after the incident which helps in evaluating the cause of the incident. The resulting video sequence of the incident is stored in the MediaMonitor database on the MediaTunnel Server, after being time and date stamped. MediaMonitor software, the operations interface of the system, is installed on both the server and the analysers. It serves as a video clip management interface and a system supervision tool for traffic operators and at the same time allows administrators to do system and interface configuration and maintenance. Video sequences over several weeks can be stored in it. Video sequences can be sorted and filtered by date, camera number, type of incident and comment in order to easily retrieve them.

For a critical system such as a traffic incident detection system, a highly desirable characteristic is redundancy – provision of backup systems in case the main system becomes inoperative due to some reason. MediaTunnel achieves server redundancy by connecting system components such as the analysers, workstations and traffic management software to two servers, thus ensuring that the video detection process continues if one server goes down. Similarly, analyser redundancy is achieved by doubling each analyser with a backup analyser.

Besides traffic incident detection, the system provides useful optional data for each traffic lane or for a group of lanes: vehicle count in lane(s), average speed of traffic in the lane(s) and travel time for a particular stretch of the road. It can also calculate advanced traffic data such as the average occupancy on a road or on different lanes of the road, and is able to classify vehicles into three classes based on their length. A TCP-based protocol can be used to supply the data to a Traffic Management Centre.

One of the problems in automatic traffic incident detection is the high rate of false alarms due to environmental factors such as heavy rain, falling snow at the entrance of the tunnel, and shadows on the road. To reduce the false alarm rate, besides tracking and trajectory analysis, the MediaTunnel algorithm uses techniques of superposition to analyse movement patterns and shapes. It also uses dynamic shaping of the reference background image to identify permanent shadows and other obstacles or events which hamper incident detection. It also applies dynamic filters upon environmental conditions such as rain, and snow which interfere with the incident detection.

Says Rodrigues, “The most significant and impressive improvement in the process is the period of response to an incident. We can now detect an incident in about 20 seconds, see how critical the incident is, and then send appropriate rescue team with the proper equipment. The impact of the incident on the traffic flow is minimised. A few months ago we had a mini-van incident on the central lane during the afternoon rush hour. After the incident was detected and analysed, we saw that no person was seriously injured. We blocked the central lane and sent in a rescue team which cleared the lane in about ten minutes.”

Share with: