ANPR-MMR & WIM for detection of overloaded vehicles

Transcription

1 ANPR-MMR & WIM for detection of overloaded vehicles Leszek Rafalski Instytut Badawczy Dróg i Mostów ul. Instytutowa 1, Warszawa, Poland Michał Karkowski Instytut Badawczy Dróg i Mostów ul. Instytutowa 1, Warszawa, Poland Cezary Dołęga. Neurosoft Sp. z o.o. ul. Robotnicza 72, Wroclaw, Poland Janusz Wrobel Neurosoft Sp. z o.o. ul. Robotnicza 72, Wroclaw, Poland Abstract Overloaded vehicles cause considerable damage to the roads and therefore reduce the length of their usability. They are also responsible for dangerous traffic accidents. Standard WIM solutions are designed to capture axle weight and gross vehicle weight. These parameters are used to detect the vehicles with exceed allowed axle or gross vehicle weight. This allows detection of overloaded trucks but not overloaded vans, which are also responsible for the road damage and road accidents. This document describes solution which is an combination of video recognition and weight in motion solution which allows the detection of overloaded vehicles. Keywords: overloaded truck and van, ANPR, MMR, WIM, video detection.

2 1. Introduction Vehicles which are loaded above the maximum authorized weight cause not only considerable damage to the roads and therefore reduce the length of their usability but they are also responsible for dangerous traffic accidents. Standard WIM solutions are designed to capture axle weight and gross vehicle weight. These parameters are used to detect the vehicles with exceed allowed axle or gross vehicle weight. This allows detection of overloaded trucks but not overloaded vans or other vehicles, which are also responsible for the road damage and road accidents. One of the most tragic examples of accidents caused by an overloaded van, was an accident that took place in Poland on the national road No. 707 on the outskirts of Nowe Miasto on the Pilica river. In this crash a Volkswagen bus collided with a truck. The accident killed 18 people, who were all van passengers. This case shows how thoughtless drivers can be. Picture 1 Tragical car accident nearly Nowe Miasto The solution described below prevents damage to the roads caused by any kind of vehicle and also prevents such dangerous accidents. The solution uses two detection subsystems: Weigh in motion and Video identification which allows number plates, make and model recognition. The implemented solution uses the following components: Weigh in motion software and electronic equipment from Traffic Data Systems GmbH (Germany), LINEAS quartz sensors from Kistler Instrumente AG (Swiss) video cameras from Bosch AG ANPR-MMR engine (automatic number plates, make and model recognition) and additional hardware and software from IBDiM and Neurosoft Sp. z o.o. (both Poland) Supervising software from Neurosoft Sp. z o.o. The main idea of the solution based on comparing two sources of data - real weight of the vehicle and legally allowed maximum vehicle weight (gross vehicle weight).

3 2. WIM subsystem 2.1 Quartz sensors The main component of the system is a module for dynamic weighing which was constructed on the basis of specialized pressure Kistler Lineas Quartz Sensors (type 9195F) dedicated for calculation of the WIM type pressure. This sort of pressure sensor is characterized by very impressive metrological parameters, e.g. high precision in a full range measurement, low sensitivity to temperature changes (less than 0.02%degrees Celsius, excellent linearity in a full range measurement (deformation %FSO less than +-2). In addition during the mounting process these sensors require minor interference in a road surface which means that the damage to its structure is very minor. For this reason these sensors can be successfully mounted in the concrete as well as the asphalt pavements. Load bearing coat (can be ground) Elastic material (foam) Quartz sensing element Special alloy profile Sand- epoxy grouting Picture 2 Lineas Quartz Sensor In one measuring point four sensors with dimensions 5x175cm are used, aligned into two measuring lines covering the entire width of the lane, which is 350cm. During the passage over the measuring point each wheel is weighed twice. Based on the double measurement of the pressure of each wheel, after taking into account the speed of the vehicle measured by the classification module, the weight (force of the pressure), for each axle is determined, as well as the total weight of the vehicle. Results of the analysis are available within less than a second from the moment of vehicle s departure from the measurement point.

4 Picture 3 Installation of 2 rows of lineas quartz sensors Thanks to good quality of pavement, special sensor construction as well as using double set of sensors the measurement system achieves the best available accuracy (A class) in accordance to COST 323 Weigh-in-Motion of Road Vehicles, Final Report, Appendix 1, European WIM Specification (version 3.0, September 1999). This means that in optimal weather conditions the system is able to rate the total vehicle s weight with a tolerance +-5%, weight of group of axles with a tolerance of +-7%, weight of axles within the group with a tolerance of +-10% and weight of a single, ungrouped axle with a tolerance of +-8. These tolerances also concern vehicles with a total weight above 3.5t and axle pressure above it. 2.2 WIM-Computer The WIM subsystem aggregates the loads of wheels, axles, axle groups and vehicles weight. To improve the accuracy, an average value is calculated from the two rows of sensors provided. The two weight measurements are carried out independently. The inductive loop classifier is required for segmenting the stream of vehicles. An appropriate signal from the inductive loop classifier indicates the end of the vehicle, which would otherwise not be recognised in the pulse train from the row of sensors, especially in heavy traffic. The inductive loop classifier also provides the speed required for the weight calculation as well as classifying vehicles into 8+1, or 13+1 vehicle classes. These provide data related to the wheel, axle, axle group and gross weight of the vehicle, from which a maximum load is determined. Very important component of the vehicle measuring system is a vehicle classification module using induction loops mounted in the road surface, between the weight measuring sensors. This module provides the following information: vehicle class,

5 exact vehicle speed, total vehicle length and spacing between the axles, the total length of vehicle s passage through the measurement point (in milliseconds), distance between any two vehicles. The WIM-System uses the best available module for vehicle s loop classification TDS 821R made by Traffic Data Systems GmbH from Germany. Device TDS821R is connected to the loop system in a configuration TLS type 2 this configuration is recommended for measurement points located on national roads, where vehicles travel at a moderate speed. This device, as the only one in the world has class A1 quality certificate according to German standards TLS (2002) - Technische Lieferbedingungen für Streckenstationen, Ausgabe Among others this allows to ensure accuracy of vehicle detection at a level higher than 99% of all vehicles driving through the measurement point. Picture 4 TDS Modul for vehicle classification TDS 821R module is located in a case hung on a gantry, in a distance of 20 meters from a loop. Pressure sensors are also connected to this module measurement data from the pressure sensors is analyzed and consolidated and then sent to the computational computer located at the permanent weighing station by Ethernet network (TSP/IP). TDS 821R module classifies vehicles according to the following basic categories (in accordance with TLS (2002) 8+1 specification: motorcycle car van

6 car/van with a trailer truck truck with a trailer tractor with a trailer bus unclassified vehicle Picture 5 Used vehicle classification In addition to basic categories a vehicle s shape classification is included. This classification indicates the way of axles grouping, including division to a vehicle and a trailer. As a result of this the system assigns the studied vehicle into one of 56 specific classes defining its shape. 3. Video identification subsystem The video subsystem is responsible for the automatic identification of the vehicles based on their images generated by a video camera. Within the identification process for each vehicle the system determines the following parameters: content of the registration plate and the country of origin affiliation to one of the vehicle classes (motorcycle, car, van, truck, bus) and make, model and colour of the vehicle (MMR). The system monitors all road traffic vehicles driving with the speed of up to 250 km/h. The core of the system is the software based on the neural network technology which uses DSP techniques (NC Engine). Thanks to applying modern multi-core processors with a high computing power, there is no need to use any specialized equipment for the image processing. The primary software platform is the operating system Linux (32 or 64 bites). For the registration of the passing vehicle the system does not require additional elements which would trigger taking a photo (triggering). The identification is carried out only on the basis of the analysis of the photo sequences supplied by a video camera. Based on the generated photos the system detects the presence of the vehicle and then determines all the parameters mentioned above. In each measurement point, on a gantry located around 20 meters away from the sensors, two video cameras are mounted: Overview video camera which registers a view of an entire road, including

7 an overview of the controlled vehicle, Measurement video camera which captures the front view of the vehicle at the time of driving onto the weight sensors, this camera s task is capturing the content of the number plate and make & model of vehicle, which needs to be automatically recognized. Measurement video camera is equipped with a special infrared light radiator allowing receiving clear images of registration plates from vehicles driving at a high speed in the dark. This radiator creates a light completely invisible to the human eye (wave length - 940nm), with 30 degrees beam width, so for this reason its light doesn t pose a threat to be mistakenly taken for a red traffic light above the traffic lane. Video camera is mounted in a special case (with ventilation and heating) allowing it to function accurately even in extreme weather conditions. Picture 6 Camera for number plates recognition (ANPR) Picture 7 Picture from the ANPR camera Moreover in all measurement video cameras special shields are mounted protecting them from getting soiled, which would affect the ability of the number plate recognition. These sorts of shields are also used in case of placing a camera directly within the traffic lane and their main task is eliminating possibility of a camera lens getting soiled as well as protection from the adverse sunlight. Both of these factors (getting soiled or blinded by the sunlight) could have a negative impact on the quality of ANPR-MMR recognition system. Depending on the vehicle speed and the way the camera is mounted, three to twenty images can be generated for each vehicle. If the camera is mounted in a standard way and a car is driving with the speed of 90 kph, the registration plate will be visible on 7-9 photos.

8 Picture 8 Camera for overview picture Picture 9 Picture from the overview camera For all passing vehicles the system finds and identifies the number plate and at the same time a special prepared neural network also recognizes a make and a model. The whole process of identification takes less than 40 ms. Several dozen of photos are generated for one vehicle depending on its speed, time of the day and season. The procedure of identification is performed for each photo, the result is classified, and finally the frame with the best identification result is chosen. Additional camera installed on the gantry delivers an overview picture with the site view of the vehicle. Picture 10 Region of interest for MMR tuck Picture 11 Region of interest for MMR - car The quality of identification of the registration number depends on the lighting conditions and the scene geometry defined as a number of parameters: perspective rate, zoom, turning angle. In good lighting conditions and with optimal adjustment the system shows the following effectiveness for a regular traffic: 99% correctly recognized registration plates 98% correctly recognized types 95% correctly recognized makes 80% correctly recognized colours.

9 4. Supervising unit All the data from the video identification unit and from WIM unit are time synchronized. The supervising unit prepares data package which contains following information s: the data delivered from a dynamic scale, recognized number plates, recognized make and model pictures of the vehicles and sends those to the central system. Additionally the system compares legally allowed maximum vehicle weight determined on a make and model with a real total vehicle weight delivered by the WIM system. If the real weight exceeds the allowed weight, the system generates alerts by SMS, SNMP, or direct via the warning screen in the user interface application. 4.1 Consolidation and archiving the measurements Responsibility for joining all types of measurements from different devices falls onto a module, which consolidates all the results and which is a part of NeuroCar 2.0 Terminal software. This module receives data from recognition module and from weighing and vehicle classification module. After receiving this data it is combined into integrated information about a vehicle that way for each vehicle there is one set of measurement data saved as an XML file. To this XML file there is also added exact information about localization of the measurement point as well as precisely defined time of capturing the vehicle (date, hour, minute and millisecond). To this measurement file additionally photos of the vehicle are added one image from the measurement camera (front view of the car) and appropriate image from the overview camera (overview of the car). After creating data package containing description of single vehicle the system analyzes content of this data and determines whether a specific vehicle has committed an offense. If that is the case the system forwards the entire package of data to a local archive offence register. Meanwhile a signal for presentation module is generated, that another incident of offense has been recorded. Afterwards the data package is forwarded to the data transmission module and prepared to be sent. One of the responsibilities of the data consolidation module is to maintain accurate time synchronization of all devices delivering measurement data milliseconds synchronization accuracy is required. For this purpose an NTP protocol is used (computational computer WORK is a time server) and the time synchronization is forced every few minutes. 4.2 Presentation module The main responsibility of the system is current control of passing trucks and catching these, which exceed the allowed maximum weight (42t), exceed the allowed axle pressure (11,5t). Additionally all vehicles, also f.e. vans which exceed allowed maximum weight, or which exceed the allowed speed or length can also be recorded in the offence register. Access to locally gathered information is possible thanks to the data presentation module which is a part of NeuroCar 2.0 Terminal software.

10 Picture 12 Information about truck with exceeded axl weight This module provides data thanks to www server built-in the system, which is why people who use the system can plug into it from any computer connected to the local network. To connect a user with a computer any www. search engine able to run Java applets (MS Internet Explorer, Mozilla Firefox, Google Chrome, Apple Safari, Opera) will be required. After opening a website in a website s search engine the user has access to all functions of the system, including: preview of the system s condition, preview of the last 10 recorded vehicles, preview of the last 10 vehicles recorded in the offence register, system configuration. Each of the websites presenting data is refreshed automatically (AJAX technology) immediately after new incident appears, for example after recording another overweighed vehicle. In case of websites displaying the offense register, each new offence generates a warning signal. 4.3 Data transmission module The responsibility of the data transmission module is transmission of the measurement data to the central repository. Moreover all the administrative and maintenance work can be carried out remotely. The module is a part of Neurosoft s NeuroCar 2.0 Terminal system. For the data transmission system uses different communication platform: GPRS/EDGE/UMTS/CDMA WiFi Ethernet WiMax.

11 All communication channels provide a safe connection of the system with a public Internet network but without an external IP address. Responsibility for maintaining the connection with the central repository is held by a data transmission program module operating on a communication computer located in the control point at the permanent weighing station. This module has the VPN connection with the centre what assures full data protection despite using the public Internet network. VPN connection is restored automatically in case of temporary problems with the radio connection. Additionally an algorithm for automatic switching of routing paths is applied in case there is a longer failure of the main transmission channel (Wi-Fi). In a circumstance of longer loss in communication all the measurement data are kept in a communication computer until the connection is restored. After combining all information from all sensors and CCTV cameras a passage of each vehicle is recorded as an XML file (datagram) containing all information from sensors including two photos images of the recorded vehicle from the L1 measurement camera and image from the V1 overview camera. pl-dk4woladebinska-tar-l xml pl-dk4woladebinska-tar-l jpg pl-dk4woladebinska-tar-v jpg The entire system is based on thesis of the original design, according to which the set of measurements for one vehicle is stored in the archive in separate files. Names of these files contain a prefix, which defines country and name of location (PL=Poland, dk4=national Road No.4, woladebinska=location Wola Dębińska), infix defining direction and lane number or overview camera number (tar=tarnów direction, l1=free lane, v1=first overview camera) as well as a suffix defining precise moment of taking the measurement ( = 23 March 2010, = 09:12 and 49 seconds, 895=895 ms). In the offense register all above files are recorded. In a case of statistical data transmitted to the central repository, XML files and image from L1 camera are sent. Sample XML datagram for vehicle s single passage has a form of: <?xml version="1.0" encoding="utf-8"?> <vehicletrace> <location> <country>pl</country> <name>dk4woladebinska</name> <direction>tar</direction> <lane>l1</lane> <latitude> </latitude> <longitude> </longitude> </location> <timestamp> <date> </date> <time>09:12:49</time>

12 <ms>895</ms> </timestamp> <exdata> <data source="camera" name="plate1"> <value name="level">75</value> <value name="country">pl</value> <value name="symbol">wp XXXXX</value> <value name="prefix">wp</value> <value name="position">299,413;299,444;412,439;412,408</value> <value name="type">white zwykła I</value> </data> <data source="camera" name="vehicle"> <value name="direction">1</value> <value name="color">none</value> <value name="type">none</value> <value name="height">0.00</value> <value name="width">0.00</value> <value name="length">0.00</value> <value name="speed">0.00</value> </data> <data source="scale" name="vehicle"> <value name="speed">59</value> <value name="length">15.100</value> <value name="direction">1</value> <value name="weight">46.986</value> <value name="avc">9</value> <value name="lane">1</value> <value name="sclass">106</value> <value name="lwheel4">4.235</value> <value name="lwheel5">4.263</value> <value name="lwheel6">3.924</value> <value name="lwheel1">3.624</value> <value name="lwheel2">2.048</value> <value name="lwheel3">5.104</value> <value name="axis4">8.789</value> <value name="axis5">9.054</value> <value name="axis6">8.543</value> <value name="axis1">6.985</value> <value name="axis2">3.842</value> <value name="axis3">9.773</value> <value name="dist5">1.320</value> <value name="dist4">1.420</value> <value name="dist1">2.890</value> <value name="dist3">4.770</value> <value name="dist2">1.220</value> <value name="rwheel2">1.794</value> <value name="rwheel3">4.669</value> <value name="rwheel1">3.361</value> <value name="rwheel6">4.619</value> <value name="rwheel4">4.553</value>

13 <value name="rwheel5">4.791</value> <value name="maxaxisweight">9.773</value> </data> </exdata> </vehicletrace> 5. References (2010), Tragic accident nearly Nowe Miasto Kistler Instrumente AG (2005), Lineas WIM technology for traffic on the go Traffic Data Systems GmbH (2010), Weigh in motion preselection WIM-P Neurosoft Sp. z o.o., (2010), ITS Congress in Poland Automatic identification of overloaded and oversized vehicles