Abstract Information about vehicles on the road is very important for the maintenance of traffic control in current complex traffic condition. Images of vehicles are captured by vehicle-directed cameras. This paper proposes a new vehicle tracking mechanism using license plate recognition technology, which is essential to having information about vehicles on the roads. The proposed method is a real-time processing system using multistep image processing, as well as recognition and tracking processes from 2D and 3D images. The experimental results of real environmental images in recognition and tracking using the proposed method are shown.
licence plate recognition software 11
Identifying a vehicle of interest is one thing, but finding it is another. With Vigilant VehicleManager license plate recognition software, start your search using effortless hot list creation and sharing paired with various options for receiving alerts when a vehicle is sighted. Additionally, expand your search by identifying associate vehicles or take a proactive approach using predictive analytics to determine where a vehicle might be in the future.
Number or license plate recognition has become an essential technology for traffic and security applications. Providing access control at any organization or academic institution improves the level of security. However, providing security personnel to manually control the access of vehicles at an academic institution is costly, time-consuming, and to a limited extent, error prone. This study investigated the use of an automated vehicle tracking system, incorporating experimental computer vision techniques for license plate recognition that runs in real-time to provide access control for vehicles and provide increased security for an academic institution. A vehicle monitoring framework was designed by using various technologies and experimenting with different camera angles. In addition, the effect of environmental changes on the accuracy of the optical character recognition application was assessed. The Design Science Research methodology was followed to develop the vehicle monitoring framework artifact. Image enhancement algorithms were tested, and the most viable options were evaluated and implemented. Optimal operating criteria that were established for the vehicle monitoring framework achieved a 96% success rate. The results indicate that a cost-effective solution could be provided by using an existing camera infrastructure at an academic institution and suitable license plate recognition software technologies, algorithms, and different camera angles.
Chen, L., Cui, L., Huang, R. and Ren, Z. (2016), "Bio-inspired neural network with application to license plate recognition: hysteretic ELM approach", Assembly Automation, Vol. 36 No. 2, pp. 172-178. -11-2015-105
Automatic number plate recognition (ANPR) systems are becoming vital for safety and security purposes. Typical ANPR systems are based on three stages: number plate localization (NPL), character segmentation (CS), and optical character recognition (OCR). Recently, high definition (HD) cameras have been used to improve their recognition rates. In this paper, four algorithms are proposed for the OCR stage of a real-time HD ANPR system. The proposed algorithms are based on feature extraction (vector crossing, zoning, combined zoning, and vector crossing) and template matching techniques. All proposed algorithms have been implemented using MATLAB as a proof of concept and the best one has been selected for hardware implementation using a heterogeneous system on chip (SoC) platform. The selected platform is the Xilinx Zynq-7000 All Programmable SoC, which consists of an ARM processor and programmable logic. Obtained hardware implementation results have shown that the proposed system can recognize one character in 0.63 ms, with an accuracy of 99.5% while utilizing around 6% of the programmable logic resources. In addition, the use of the heterogenous SoC consumes 36 W which is equivalent to saving around 80% of the energy consumed by the PC used in this work, whereas it is smaller in size by 95%.
Modern cities are implementing intelligent transportation systems (ITSs) as they are an essential part of the infrastructure especially with the increase of population and number of vehicles. The system that identifies vehicles by recognizing their number plates (NPs) is known as an automatic number plate recognition (ANPR) system and it is a part of ITS. It does not require any pre-installed equipment in the vehicle. They are used for several purposes including car park management, law enforcement, counter-terrorism and security, tolling, traffic monitoring, vehicles tracing, and cloning prevention [1,2,3,4]. Car cloning is a car identity theft that is achieved using a false NP.
OCR algorithms are important and widely used to translate the content of scanned images into encoded text. Their usage ranges from tasks involving translating hand writing on notebooks, parcels, and checks to simpler character recognition tasks since the text in the captured image has a uniform font and was taken under good lighting conditions. For ANPR systems, character recognition is relatively less complex because fonts are usually uniform and NPs in some countries include only numeral characters. However, a challenge lies in dealing with the different conditions of these NPs, such as being under sunlight or shadow, dirty, rotated, or having damaged paint [9]. Good OCR algorithms must be able to handle these conditions efficiently. Furthermore, as NPs contain a string of several characters, one mistake is enough to wrongly render the detected plate. Thus, the performance of the full ANPR system is affected by this stage performance. In general, OCR algorithms fall into four categories: feature extraction techniques [10,11,12], template matching or correlation [13], statistical classifiers [13,14,15,16], and artificial neural networks (ANN) [5, 17]. The performance of the hardware implementations of a given algorithm differs by its ability to be pipelined and parallelized, and by the amount and type of calculations performed by these parallel blocks.
In [13], two methods were used to perform OCR. The first uses template matching where it correlates the image of the unknown character with a given set of templates to identify the character. The second one depends on support vector machine (SVM) classifiers that were trained by using feature extraction techniques to identify the input character. In [14], SVMs were used for character classification. Segmented characters from the previous stage were resized to a common size. The feature vectors consist of direct pixel values. A combination of OneVsAll SVMs and Tree-like structure classifiers were used. This was implemented on a TI C64 fixed-point DSP platform that classified one character in 2.88 ms, with a recognition rate of 94%. The proposed OCR stage in [15] is based on a simple nearest neighbor classifier.
As mentioned earlier, a typical ANPR system consists of three stages. The first stage is the HD NPL and its proposed implementation by the authors consists of three main operations: image resizing, morphological operations, and connected component analysis (CCA). The three operations have been implemented on the PS and PL units to reduce the processing time and take advantage of the pipelining in the PL. The bilinear image resizing method, and the morphological operations are implemented using the PL, whereas the CCA is executed using the PS. For the second stage proposed implementation, the CS, there are three operations: adaptive thresholding, morphological operations, and CCA. Similar to the HD NPL, the adaptive thresholding and morphological operations has been implemented using the PL, whereas the CCA is executed by the PS. For both stages, the open source computer vision (OpenCV) library is used to perform the CCA. Moreover, the segmented characters by the CS stage are resized using the PS before identifying the characters in the OCR stage. The MATLAB implementation to process one plate using the proposed HD NPL and CS takes 32.84 and 16.4 ms respectively. However, after exploiting the parallelism in the selected hardware platform, the times dropped to 16.17 and 0.59 ms respectively. The recognition rates for the HD NPL is 98% where the CS achieved 98.7%. The two stages utilize around 12% of the available resources in the selected platform. More details about the proposed HD NLP and CS stages by the authors can be found in [18, 19].
The optimization of the fractional part shown in Eq. (5) is more difficult than the integer part since it affects the recognition rate of the stage. Hence, the aim of this optimization is to maintain the recognition rate of the stage when compared to the software implementation using MATLAB while achieving better performance estimates.
For the FP (30, 11) fixed-point variable, the absolute error per sample does not reach the threshold as shown in Fig. 11. However, Fig. 12 indicates that its maximum absolute error is close to the threshold. Meanwhile, the number of samples used in this testing process is small; hence, there may be a case where the absolute error exceeds the threshold. Consequently, to be confident that does not happen, FP (31, 11) fixed-point variable is selected. Afterwards, additional nine samples are used to further test the algorithm using FP (31, 11) fixed-point variable. Eight samples are selected randomly where the ninth sample represents the worst case scenario of the testing set. If every testing set image is matched with the ten templates, there will be ten coefficients for each image. The worst case is when the difference between the highest two correlation coefficients out of the ten is the smallest. The image that has the smallest difference is the most likely affected by the introduced errors of the fixed-point variables. This image is found using MATLAB. For the 29 samples, the recognized characters by the OCR hardware implementation using FP (31, 11) fixed-point variables are the same as the MATLAB implementation; hence, both implementations have the same 99.5% recognition rate. 2ff7e9595c
Comments