1.26k likes | 1.27k Views
LPR Products and Services for Electronic Toll Collection. Include table of contents slide. Industry Requirements. Tolling Industry has called for improvements in license plate recognition system performance: Demand for reliable and complete capture of license plate data
E N D
Industry Requirements Tolling Industry has called for improvements in license plate recognition system performance: • Demand for reliable and complete capture of license plate data from all vehicles, including: • Plate type • State/Province or country of origin • License plate numbers/characters • All captured images must be accurately converted to usable information
Perceptics Imaging Systems Perceptics LPR Systems have been designed and optimized for the Tolling Industry • Engineered for accuracy, understanding the goals of a tolling agency: • Efficient revenue collection • Decreasing the costs of daily toll collection • Success of these goals depends on achieving the most accurate license plate data results possible
Perceptics, LLC Corporate Information • US Corporate Headquarters, Farragut, Tennessee (Knoxville) • 35+ Years in Business, Primary Focus Border Security • 1980 – Began Development of LPR with US Customs and Border Protection • 2014 – Launch of Seventh Generation of Products (G7 Series) • 2015 – Grand Opening of new Corporate Headquarters • Continual Investments in People, Products & Position • Mission Critical Imaging Technology Key Markets • Border Security • Highway & City Surveillance • Commercial Vehicle Enforcement • Electronic Toll Collection Geography: North America
Timeline slide 35+ Years in Business, Primary Focus Border Security 1980 – Began Development of LPR with US Customs and Border Protection 2014 – Launch of Seventh Generation of Products (G7 Series) 2015 – Grand Opening of new Corporate Headquarters
ETC System Block Diagram LPR-ETC In-LaneLPR Equipment In-LaneOCR Engine(optional)
Attach Rate / Error Rate 95% / 1% Perceptics Imagers (LPR-ETC) 98% / 0.5% Perceptics Imagers paired with Perceptics OCR Server 85% / 1% Perceptics OCR Server with Third Party Imagers meeting our minimal image quality specs. Accuracy
Capabilities • Automatically detects and counts every vehicle • Reads the full license plate number and jurisdiction of origin with better than 99% accuracy • Both retro reflective and non-retro reflective license plates • Omni font OCR algorithms are scale and rotation tolerant • Recognizes stacked and half-sized characters • Plate results are transmitted in less than 1.5 seconds (2,400 vehicles per hour) • 24/7 operation in all weather conditions
Capabilities (cont.) • Captures vehicles traveling up to 120 mph (200 kph) • Recognizes passenger and commercial license plates • 15 ft (4.6 m) horizontal field of view • Can locate and identify multiple plates on a single vehicle • -40°F to 158°F (-40°C to 70°C) operating temperature range • Easy to use browser-based interface for remote status, software updates, and system configuration
Two types of license plate materials: • Retro Reflective • Non-Retro Reflective • Some materials have glass beads added to create a retro reflective surface, which means the majority of light is reflected back towards its source.
Multiple Plates per Vehicle AZ MX MX CA NL MX CA MX
LPR-ETCOpenRoad LPR 5MP System includes: • One Imager (IMG750) • One Stroboscopic Illuminator (IL5470) [Optional Baffled Stroboscopic Illuminator (IL5470B)] • Embedded System and OCR Software unique to the electronic toll collection system application. Physical Characteristics: • Imager Dimensions: 8.4”W x 7”H x 12”D (213mm x 178mm x 305mm) • Illuminator Dimensions: 10.3”W x 9.1”H x 19.1”D (262mm x 231mm x 485mm) [Baffled Illuminator Dimensions: 10.3”W x 9.1”H x 25.1”D (262mm x 231mm x 638mm)] • Imager Weight: 11.5 lbs (5.2 kg) • Illuminator Weight: 24.5 lbs (11.1 kg) [Baffled Illuminator Weight: 30 lbs (13.6 kg)] • Material: Powder-coated Aluminum Environmental: • Operating Temp: -40°F to 158°F (-40°C to 70°C) • Humidity: 10%~90% (non-condensing) • Imager Ingress Protection: IP65 • Illuminator Ingress Protection: IP54 Processor: • i.MX53 1 GHz ARM Cortex-A8; 32K L1 instruction/data caches; 256K byte L2 unified instruction/data cache • Neon coprocessor (SIMD) • 1 GB DDR-3 800MHz memory • 16 GB eMMC 4.4 onboard Connectivity: • 10/100 Mbps Ethernet port • Two High-Speed USB 2.0 host ports (Internal) • Micro USB 2.0 device port (Internal) • UART Console port (Internal) • uSD card interface (Internal) • SATA II interface (Internal)
LPR-ETCOpenRoad LPR 5MP (cont.) Imager Sensor Specifications: • Sensor Size: 2576 x 2048 active pixels • Sensor Type: Quad Super eXtended Graphics Array (QSXGA) CMOS • Sensor Readout: Random Programmable Region of Interest (ROI), Global Shutter • Optical Size: 1 inch • Pixel Size: 4.8 µm x 4.8 µm • Pixel Bit Depth: 10-bits • High Dynamic Range (HDR) • On-chip Fixed Pattern Noise (FPN) Correction • Automatic Exposure Control (AEC) • Serial Peripheral Interface (SPI) Optics: • Lens Options: from 16mm to 50mm available • Operating Range: 12 ft (3.7m) minimum, 75 ft (23m) maximum, depending on lens option and lane geometry. • Field of View: 10.5 ft (3.2m) – 21.5 ft (6.6m), depending on lens option and distance to target. Illumination: • Light Source: Xenon gas discharge lamp • Illuminator Output: 59 Joules • Recharge Time: 200 ms • Lamp Life: 10,000,000 flashes • Wavelength: depends on optical filter used Electrical: • Input Voltage: 120VAC, 150W (continuous) • Maximum Input Current: 8.75A (during charge cycle)
LPR-ETCOpenRoad LPR 5MP (cont.) Sensor Spectral Response: Illuminator Spectral Power Distribution:
Typical Performance • >99% ENTIRE license plate read accuracy • Vehicle detection is independent of condition of license plate(>99% image capture accuracy) • >99% state ID • >99% character identification • >95% attach rate • Images have high contrast and no motion blur
System Operation The Perceptics LPR is designed to automatically: • respond to a trigger indicating a vehicle in its field of view, • capture a digital image of the front (and/or rear) of the vehicle, • locate any license plate(s) in the field of view, • read the alphanumeric number, • identify the state or country of issue, and • output the resulting vehicle identification number
Vehicle Detection • Sensors detect when the front (or rear) of a vehicle is in the field of view of the LPR • Sensing options • Inductive loops • Lasers • Thru-beam photoelectric sensors
Image Acquisition Vehicle image is captured using a stroboscopic illuminator that is synchronized with a ultra-high resolution CMOS imager Light intensity of strobe Intensity Camera’s shutter time
Image Acquisition Front Rear
Image Analysis • License Plate Location • Adaptive Foreground/Background Segmentation • Character Contouring • Feature Extraction (holes, bays, etc.) • Character Recognition (decision tree, neural networks) • Context Analysis and Plate Style Recognition • State Identification (StID) • Confidence Calculation • OCR Results
License Plate Location Original License Plate Region of Interest Any area of the image that contains text is carefullyexamined to locate the best license plate candidate
Adaptive Segmentation Before the characters are processed,they are isolated from the background, including state logos, and other graphics. License Plate ROI Result of segmenting ROI
Character Contouring Group pixels into blobs: Group blobs into lines:
Feature Extraction Number, location and direction of bays Number and location of holes
Character Recognition • Each character is analyzed by independent rule-based expert systems, each compiled into an optimal decision tree for efficiency of execution • Each decision tree uses a different set of features and is optimized to recognize different sets of characters • In addition, a neural network is used to compute a per character confidence factor • A voting scheme, combined with context information, is then used to determine the best read result, and an overall confidence factor
Character Recognition Decision Tree Based Shape Analysis Number of holes one two Location of hole Char = B center top bottom Char = D Char = 9 Char = 6
OCR Decision Trees OCR 1 6 Scoring OCR 2 G 6 OCR 3 6 Context
Process is Repeated per Character OCR1 Output: _ 4 7 6 6 1 8 OCR2 Output: E _ _ _ _ _ B OCR3 Output: E 4 7 6 6 1 8 Read Result: E 4 7 6 6 1 8
Context Analysis Context may be used to correct ambiguous results. For instance: Read as: E47661BMatches style: “a n nnn nn”, where n = [0-9], a = [A-Z] Results in: E476618
License Plate Style Identification • The Perceptics StID algorithm uses • State logos • Symbols • Text layout, gaps, etc. • Syntax (e.g., “Rnnnnnn”) to identify the Style of Plate (e.g. Texas Apportioned Tractor) gap
State Identification (StID) • Once identified, the Plate Style is mapped to a Plate Type (016) and a State or Province (TX) through a configuration table.
State ID / Plate Types Logo: Plate Syntax: “annnnnn” Illinois General Passenger Issue State: Type: IL 001
OCR Results front rear Read Results: IL E476618 001 100 IL E476618 001 100 State: License: Type: Conf: IL E476618 001 100 IL E476618 001 100
All Plates are Readable… Unless: • Vehicle is missing license plate • License plate not within field of view of LPR when passing through lane • View of license plate is obstructed • View of license plate is obscured • License plate is damaged • License plate is not mounted in accordance with the law
NoRead / NoPlate • If the license plate cannot be read with sufficient confidence, a special NoRead message is sent. • If the system cannot locate a license plate in the image, a special NoPlate message is sent.
Software Configuration • All LPR image processors execute the same version of firmware • Each site is optimized on a regional basis to maximize the overall correct read rate • System is trained to recognize license plate styles which occur with the most frequency at a given site • The number of states/provinces recognized varies from site to site (typically accounts for at least 98% of the total distribution of traffic)
LPR Performance • Designed to operate 24 hours a day, 7 days a week, under any weather condition (as long as visibility is not affected) • Decision tree technology provides very fast, deterministic performance that can be enhanced and maintained incrementally (no need for retraining) • Captures images of every vehicle traveling through its field of view and real-time performance means it can process up to 2,400 vehicles per hour.
LPR-ETC Output • For each vehicle passage, the following information is provided via a TCP/IP Sockets connection: • Unique transaction ID • Date and Time of transaction • License Plate String (E4600) • State ID (NC) • Plate Type (optional) • Confidence Factor (0..100) • LPR Image • Patch Images (one per result)
Confidence vs Accuracy • The attached table shows the relationship between confidence and accuracy. • In the sample shown, read results that had a confidence greater than of equal to 88 had an overall accuracy of 99.1% (0.9% error) and accounted for 92.6% of all vehicles.