There are a whopping 1.02 billion vehicles in the world and each legitimate vehicle comes with a vehicle registration plate, known as number plate or license plate for official identification purposes. All countries make registration plates mandatory for road vehicles such as cars, trucks, and motorcycles, and the registration identifier is a combination of numeric or alphanumeric ID to specifically identify the vehicle owner. There are numerous types of license plates around the world with complex combination of alphanumeric characters and due to the rising needs to identify these plates, License Plate Recognition (LPR) Technology was created. The LPR technology had undergone major progress and now it is ready to be commercialized to automate vehicle access and to deal with security issues involving registered vehicles.
License Plate Recognition, acronym LPR usually involves capturing of photographic videos or images of the plate, which then processes by specific algorithms to produce the alphanumeric text entry of the plate for record purposes. TimeTec Smart License Plate Recognition LPR is an ultramodern technology, incorporating Algorithms and Optical Character Recognition (OCR) technology as fundamental to integrate our recognition algorithm and LPR camera to convert a scanned image into a readable alphanumeric text at greatest accuracy. TimeTec trains our algorithm to cater to all types of license plates in order to maximize its recognition power.
Due to the inconsistency and complexity of license plate variances, the algorithms must be able to distinguish which part of the vehicle is actually a license plate and derive an accurate results across various disturbances along the way.
 |
|
| License Plate Localization | |
| • | A neural network based localization algorithm will locate the location of the license plate. |
| • | License plate usually consists of standard features, such as color – Black/White, White/Black, Red/White |
| • | Consists edges which come from standard patterns (Here referring to A-Z, 0-9) |
| • | Localization algorithm is trained with few ten thousand of images, to be able to recognize the pattern of license plate in an image |
| • | This trained model will then be used for license plate localization. It is one of the supervised training method |
 |
|
| Character Recognition | |
| • | Each character has unique edges and orientation |
| • | These features are used to train a classifier. (similar to neural network) |
| • | This classifier will then take character from segmentation output and classify which character it is (A-Z, 0-9) |
| • | For each connected components, classifier will perform character classification. (Exp, ABC 1234), classifier will perform classification 7 times |
| • | At the end of classification, we have 7 character results |
 |
|
| Convert Image to Grayscale | |
| • | Once the location of license plate is confirm, the region license plate will be cropped and converted into grayscale. License plate is considered as binary image, colour information is no longer useful in further analysis |
 |
|
| Character Organization | |
| • | From the segmentation result, the location of the character is known |
| • | For Malaysia license plate, it is always top to down, left to right |
| • | The output from classifier will then arrange the result based on the location of the segmentation output |
 |
|
| Character Segmentation | |
| • | The license plate usually consist of 7-8 characters (Exp: ABC 1234, WAA 1234 B). The algorithm will then segmented the character out of the license plate |
| • | It is achieved by finding connected components in a license plate |
| • | For ABC 1234, it will be having 7 connected components |
 |
|||||
| License Plate Rules | |||||
| • | Recognition process is not always 100% accurate | ||||
| • | For some license plate, or special cases, (such as B and 8, S and 5), classifier might give wrong result | ||||
| • | There is rules to correct the result in case classifier is not performing well | ||||
| • | Sample of rules: | ||||
|
|||||
 |
|
| Noise Filtering | |
| • | It is common that IP camera will pick up some noise in license plate, sometimes, it is due to the license plate itself is not clean. Many noises will appear after the segmentation process |
| • | Noise filtering will filter relatively big/small noise compare to the license plate itself. After this process, the segmentation result (after denoising) will contain only characters |
Deployment of LPR technology does not require complicated hardware installation, instead you need to invest on LPR cameras for each lane and a PC to process the information:
| ITEM | TECHNICAL SPECIFICATION |
| 2 x LPR Camera for entry and exit | ITC215-PW4I-IRLZF27135 |
 |
• 2 Megapixel Full HD AI Access LPR Camera |
| • 1/2.8 inch 2Megapixel Progressive scan CMOS | |
| • WDR, Day/Night(ICR), 3DNR,BLC, HLC | |
| • H.265& H.264 dual-stream encoding | |
| • Powerful 2.7-13.5mm motorized lens and IR light, ideal for monitor LPR distance 3-8m | |
| • IP67, and superior performance for outdoor applications |
| 2 X LED Panel | 32x16 Dot Matrix LED |
 |
• Communication : RS 232 |
| • Display data : ASCII characters | |
| • Display dimension : 320 mm x 160 mm | |
| • Display resolution : 32 x 16 pixel | |
| • Font size : 70 mm x 50 mm | |
| • Display style : 1 line rolling or 2 lines static |
| 1x Computer | |
 |
• CPU : Core i5 Intel, minimum 3.0 GHz |
| • RAM : 4GB or above | |
| • Network : Ethernet 100Mbit | |
| • Graphic adapter : AGP or PCI-Express, minimum 1024 x 768, 16-bit colors | |
| • Hard disc space : Minimum 10 GB free hard disk space available, excluding space needed for recordings | |
| • OS : Microsoft® Windows® 10 Professional (64-bit) Microsoft® Windows® 8.1 Enterprise (64-bit) Microsoft® Windows® 8.1 Pro (64-bit) |
|
| • USB port : Min 2 (one to connect to LPR gate relay board) | |
| • Serial ports : Min 2 (each connect to LED panel) | |
| 1 x LPR Gate Relay Board | |
 |
• Power input : 12VDC 3A |
| • Output : 2 x FORM C relay | |
| Software | TimeTec iSense |
LPR has to deal with vehicle velocity and the velocity vary from one vehicle to another. To implement LPR, cameras with fast shutter speed of at least 1/1000 sec is a must to cater to high speed vehicle's velocity
The use of eclipse cameras is necessary to reduce the headlight glare and reflected light from a vehicle. The eclipse camera refracts the light toward the vehicle, to produce a clear image
Infrared cameras need to be available to handle reading in areas of low light or in total darkness. A night vision camera works in complete darkness, 0 LUX, without any light because it sees in the infrared spectrum. All infrared cameras have diodes around the cameras that emit infrared light and a special chip inside the camera can capture this infrared radiation and convert the radiation into a visible picture
To produce an even better images, LPR needs to find a camera that has additional infrared illumination because the standard led on a night vision may only cover the distance of 4.5 to 9 meters. An infrared illuminator is required for longer distance around 15-20 meters and more
| • | A high accuracy rate of 95% or greater |
| • | Fast processing time |
| • | Tolerant to distortion and blur |
| • | Support high speed vehicles |
| • | Support multiple license plates formats |
| • | Supports license plates from multiple geographic regions |
| • | Can be integrated with Access control and integrated I/O modules |
| • | Can be integrated with standard CCTV systems |
| • | Can be integrated with 3rd party applications |