I do not know whether you have noticed, whether it is the parking lot charge, or the road barrier of the vehicle violation detection, has slowly realized the "unmanned" : through the camera and other image intake equipment to shoot the license plate, automatic recognition, and docking with the back-end management system, can automatically execute the vehicle parking billing, violation records and other applications. Compared with the traditional manual identification and management, automatic vehicle identification and management not only reduces the corresponding personnel cost, but also effectively improves the speed and accuracy of management, becoming an indispensable part of the intelligent transportation system.
But at the same time, if deep learning-based vehicle recognition monitoring wants to meet the requirements of accuracy and performance, it needs efficient algorithm support and a large amount of data for training. However, algorithms trained for a certain scenario often cannot fully meet the application requirements of other scenarios. Take license plate recognition as an example. The license plates of different countries and regions are different in terms of text, color and combination. Therefore, without training based on certain data, the license plate recognition algorithm is difficult to migrate to other countries and regions, which also causes the slow development of intelligent transportation system in some countries and regions with small market scale.
In addition to algorithms, computing power is also an important reason affecting the application of deep learning-based vehicle recognition system. The higher the computing power, the easier it is for vehicle recognition and detection to achieve higher efficiency, which is more suitable for harsh application scenarios, such as identifying vehicles in high-speed motion or complex scenes. The enhancement of reasoning performance helps accelerate the training of algorithms. At the same time, in order to achieve the best cost effectiveness, users often want to be able to carry more vehicle identification load with fewer nodes, which puts forward higher performance requirements.
Smart Core original motor license plate recognition system "out to sea" road
As a leading artificial intelligence technology and solution provider in China, Beijing Zhixin Yuandynamic Technology Co., LTD. *(hereinafter referred to as "Zhixin Yuandynamic"), invested by Intel in 2018, has been committed to providing users with efficient, high accuracy, rapid deployment of vehicle identification solutions. The cloud vehicle model identification platform of Zhixin Original movement can identify about 1,600 vehicle models, including the information of brand, model and age. In order to meet the needs of users in different regions of the world for license plate recognition, Zhixin original motion launched the license plate recognition solution.
In view of the defects of traditional license plate recognition schemes that require a large amount of data for training, slow deployment speed and long cycle, by using the self-developed license plate algorithm framework, the Intelligent Core original overseas license plate solution can quickly iterate under the condition of a small number of (>1K) license plate samples, and the delivery cycle of only 2-4 weeks can realize the development task of new national license plates. And the comprehensive accuracy can be as high as 90% to 95%, can meet the needs of basic applications, and with the increase of training data, this accuracy will continue to improve.
In this scheme, the intelligent core primitive motor uses the deep learning method based on convolutional neural network for vehicle model recognition, and realizes the classification reasoning optimization through the topological structures such as MobileNet* and GoogleNet*. After optimizing the license plate recognition algorithm, it is helpful to realize the rapid development and deployment of applications on the premise of small samples.
At present, Intelligent Core original motor license plate recognition solutions have been in Canada, Turkey, Singapore, Zambia, Angola, China, China, Taiwan, Thailand, India and other more than 20 countries and regions to achieve the product landing.
Intelligent Core original motion and Intel cooperation to accelerate the reasoning performance of license plate recognition applications
"Hardware and acceleration tools are very important factors to improve the reasoning performance of deep learning-based license plate recognition, and we are in ongoing exploration with Intel on both fronts." Wang Zheng, Intelligent Core original motion CTO, said, "In the deep learning solution based on CPU, the performance of CPU and technological innovation for artificial intelligence are crucial. The newly released second generation Intel ® Xeon ® scalable processor is optimized for the application load of artificial intelligence, equipped with VNNI and other technologies. We look forward to improving reasoning performance with this processor."
In addition to using the second generation Intel ® Xeon ® scalable processor, SmartCore primordials also tested the reasoning performance of the Intel ® OpenVINO™ tool suite distribution in multiple topologies such as MobileNet*, Mobilenet-V2 *, GoogleNet*, VGG-16*, etc. Testing data showed a significant performance improvement of 28.4 times in MobileNet family classification reasoning using OpenVINO™ compared to Intel optimized version Caffe.
This acceleration is possible because OpenVINO™ supports both deep learning and traditional computer vision approaches, including a deep learning deployment tool suite that enables developers to deploy trained network models on target platforms for reasoning operations.
"The use of the second generation of Intel ® Xeon ® extensible processor + OpenVINO™ in specific application examples can improve the reasoning performance in video image analysis to a very considerable degree, which not only has a huge application prospect in license plate recognition applications, but also can be applied in a wide range of intelligent transportation systems. By deploying the optimized version of the solution, users will be allowed to deploy fewer nodes while supporting more reasoning load, achieving a lower total cost of ownership (TCO), "said Zhixin, head of the original motor license Plate recognition project.
Vehicle detection and recognition will leverage the intelligent transportation market
In the huge system of intelligent transportation, the application of computer vision technology is indispensable. Whether it is comprehensive service of parking lot, road management, or comprehensive detection of road bayonet, they all rely on efficient vehicle-oriented video image analysis ability. At present, transportation industries around the world are faced with the challenge of low digitalization level. Enterprises need to connect these numbers and information, realize the commonality of information and data, and generate common value, which will bring huge opportunities for the construction of intelligent transportation system and smart city.
Computer vision-based solutions leverage enhanced deep learning neural networks to capture data in a more sophisticated manner, taking vehicle recognition oriented analytics to a whole new level. Artificial intelligence methods such as deep learning use trained algorithms to model various levels of abstraction in the data through hierarchical neural networks, which can help build complex processes such as computer vision, natural language processing and image recognition, and extract massive data information from vehicles running in the transportation system.
From Intel ® chip-based smart cameras, to Intel acceleration chip-based end-computing devices (such as network video recorders, gateways, video analytics devices, etc.), to cloud environments running training and analytics functions, Intel offers a very large portfolio of products to support AI use cases from cameras to cloud environments. The collaboration with SmartCore is part of Intel's broader plans for computer vision, which are expected to take root in more niche industries and enable the smart digital age.
