Bringing AI to public safety with advanced vehicle license plate analysis

Bringing AI to public safety with advanced vehicle license plate analysis
Governments around the world are investing more in public safety solutions, including smart technologies running algorithms based on artificial intelligence (AI) and machine learning. Public safety organizations, such as law enforcement, fire and emergency medical services are all starting to utilize AI in-vehicle; however, to be effective, in-vehicle computers must be capable of accurately processing complex, energy-hungry algorithms in challenging environments.

Despite any obstacles in connection with AI and smart technologies, speakers at the 2019 Organization for Security and Co-operation in Europe (OSCE) Annual Police Experts Meeting stated that law enforcement authorities and agencies should embrace AI and keep up with technological innovations. Even though the use of AI is relatively new for law enforcement, the OSCE emphasized its potential to increase efficiency and effectiveness. They pointed to the benefits that could be gained from using video and image analysis software, facial recognition, biometric identification.

How in-vehicle computers can enable AI for public safety

Today, public safety vehicles are starting to utilize more AI applications. For example, police cars can be outfitted with face recognition software to help identify persons registered to criminal databases. Even more often deployed are license plate recognition (LPR) algorithms, also known as automatic license plate recognition (ALPR) or automatic number plate recognition (ANPR). Law enforcement can use LPR to identify vehicles and check them against various blacklists (e.g., cars with outstanding warrants, expired registration, stolen vehicles, etc.), which they can connect to via cellular networks. 

There are major advantages to running LPR at the edge in a police vehicle. For one, it eliminates the back-and-forth communication between officers and a central command center making for more efficient and effective use of time. It also provides real-time, immediate checks, which allows officers to question drivers of blacklisted vehicles as soon as possible and solve cases faster. Utilizing LPR also makes it possible for police officers to continuously scan multiple vehicles at a time; this includes vehicles passing on the side and the opposite direction, as well as vehicles in front and parked on the side of the road.

When it comes to emergency medical services, ambulances can be equipped with in-vehicle computers that run AI applications to provide them with the fastest routes to scenes of emergency and health care facilities. It can also connect to traffic management systems, enabling traffic con-trollers to clear a path for emergency vehicles. 

However, for any of this to be possible, advanced in-vehicle computer systems are required. Running AI algorithms requires a large amount of processing power, which is made even more difficult when the size of the computer is constrained and must fit into a small space with a limited power source and work under extreme conditions.

Requirements for in-vehicle computers

Public safety applications, such as police vehicles, require stability, speed, accuracy and mission-critical reliability. In such an environment any failure could result in dangerous, life-threatening situations. Additionally, in-vehicle computers are subject to harsh, ever-changing conditions, including extreme temperatures, vibration and shock. In-vehicle computers must also be power efficient and able to deal with a fluctuating power source, since it will run off the vehicle's power supply.

Fighting against vibration

Vibration is one of the most common causes of computer system failure. In a vehicle where movement, vibration and shock are impossible to avoid, it is crucial that in-vehicle computers are rigorously tested and meet standards. 

Fanless computers are becoming more popular and are ideal for in-vehicle use since it reduces the number of moving parts within the computer, which ultimately leads to better reliability and less maintenance. Fanless computers also suck in less dust and dirt, which could help reduce future related performance issues.  

It is also important that storage can be optimized for different applications and environments. Many in-vehicle computers have flexible storage options to allow end users to choose the solution that most suits their need. For instance, the use of solid-state drives (SSD) are often preferred over standard hard drives (HDD) that have spinning, moving parts, as SSDs are more resistant to vibration than HDDs. 

Meeting relevant standards

To ensure in-vehicle computers are capable of withstanding a tough environment, in addition to basic standards, they should be compliant with more rigorous standards. For example, the United States Military Standard MIL-STD-810 was developed specifically to test equipment used for military applications, but is also used to certify commercial ruggedized products such as in-vehicle computers. The standard addresses a broad range of environmental conditions such as extreme temperature, dust and sand exposure, shock and vibration, explosive atmosphere, etc.

Running AI on in-vehicle computers with a deep learning accelerator card

In-vehicle computers need to support ultra-low power consumption and enough processing power to effectively and efficiently run advanced algorithms. This can be problematic for small in-vehicle computers that need to fit in small, tight spaces with limited power and under varying conditions. 

Support for advanced processors such as Intel's Movidius Myriad X VPU, together with the Intel Distribution of OpenVINO toolkit, can enable public safety vehicles to run AI applications, such as LPR, face recognition, pedestrian detection, etc., at the edge without consuming large amounts of power or slowing down the system.

Intel's Movidius Myriad X VPU is the latest and most advanced processor from Intel. It features a neural compute engine and 16 SHAVE cores, making it an ideal choice for on-device deep neural networks and computer vision applications. The VPU's ultra-low-power design also makes it possible to run complex algorithms without the heavy power consumption usually associated with such technologies.

Intel's OpenVINO toolkit accelerates the development of high-performance computer vision and deep learning inference into AI applications from edge to cloud. It enables deep learning on hardware accelerators and easy deployment across multiple types of Intel platforms. By using advanced processors like Intel's VPU together with Intel's OpenVINO toolkit, in-vehicle computers are able to increase the accuracy and efficiency of visual analysis at the edge.

Running AI applications such a face recognition and LPR also requires that the in-vehicle com-puter supports connection to multiple high-resolution IP video surveillance cameras. Computers should be equipped with Ethernet ports capable of effective power management, such as power-over-Ethernet (PoE) ports that budget power usage. 

Winmate's Box-IWAI in-vehicle computer brings intelligence to the edge

The latest computer offering from Winmate, Taiwan-based manufacturer of rugged technologies, is a fanless embedded VPU-based in-vehicle computer — Box-IWAI. The in-vehicle computer is easy to set up and best suited for smart traffic AI applications, as well as use in public safety vehicles such as police cars.

The Box-IWAI is a rugged fanless embedded in-vehicle computer powered by the 8th generation Intel Core processor. It is fitted with four PoE ports with a total power budget of 60 Watts for up to four IP video surveillance cameras, and supports ONVIF IP cameras and DVR/NVRs. It supports GPS and G-sensor, has an operating temperature range of -20C to 60C, and has a flexible I/O window slot to meet the requirements of various applications. The Box-IWAI is E-Mark, CC, FCC and RoHS certified — it is also MIL-STD-810G anti-shock and vibration resistant.

From an AI perspective, Winmate's Box-IWAI supports up to eight Intel Movidius Myriad X VPUs. Support for Intel’s latest VPU opens the Box-IWAI up to an entire range of new and more advanced AI applications. The AI computing does not require calibration and can be used immediately upon installation. 

Winmate optimized for AI

As 5G technology slowly rolls out, the need for fast, accurate video transmission will be critical for advanced visual analysis. This is where Intel's VPU excels. Intel’s VPU provides flexible image processing and hardware-based encode for up to 4K video resolution, as well as new enhanced vision accelerators. Although the use of more advanced processors like Intel's VPU generates more heat, Winmate's fanless design has been optimized to dissipate any excess heat generated. 

Winmate's computer solution also comes pre-installed with an inference system and pre-trained models such as LPR, vehicle classification, vehicle counting and pedestrian detection. With Intel Distribution of OpenVINO toolkit integrated, customers can develop and optimize their own computer vision applications with the OpenCV library or OpenVX API without spending extra time or money deploying presets, private clouds or big data imaging solutions. Customers can also use Winmate's 3-Party-AI software API to import existing trained models from Caffe, TensorFlow, Apache MXNet and ONNX and deploy them to edge devices.

Winmate points out that in order for AI object recognition engines to learn, it requires tens of thousands of tagged images; a time-consuming process. Therefore, it needs a significant storage to put all the images in the local place. Winmate's Box-IWAI deep learning object recognition technology, eliminates this process by including pre-trained models that include object image tags, image recognition and accurate verification, eliminating the need to buy expensive cameras with built-in analysis systems. 

Winmate for law enforcement vehicles

Winmate has partnered with Taiwan-based AlphaInfo to include its AI software framework into the Box-IWAI — AlphaInfo specializes in the development and application of AI, the Internet of Things (IoT), big data, data mining and cloud computing. Optimized by Intel’s OpenVINO toolkit, AlphaInfo’s AI-based LPR system is available to law enforcement pre-loaded into Winmate's Box-IWAI. 

With AlphaInfo's LPR software, law enforcement can simultaneously grab high-resolution images of license plates from passing vehicles, as well each vehicle's color and make. Vehicle identification can be made in fractions of a second and then compared to relevant databases in the cloud. The LPR can even identify vehicles passing at high speeds of 110kmh and above, and coming from multiple directions. Additionally, the intelligent LPR algorithm uses deep learning to constantly get better and learn. Unlike traditional systems that require two separate cameras for video surveillance and LPR, Winmate's AI software can be integrated with the IP camera thereby reducing overall equipment costs.  

When connected to high-resolution IP surveillance cameras, the Box-IWAI acts similarly to a mo-bile NVR inside a police vehicle. Footage from the cameras can be processed locally in real-time with the LPR software and checked against blacklist databases, all without having to be sent to a central control center. 

Enhanced processing power for AI-enabled future

As the demand for smarter systems and solutions continues to grow across various in-vehicle applications, the need for advanced in-vehicle computers with increased processing power able to run AI algorithms will only intensify. In-vehicle computers will need to support the most advanced processors and be optimally designed for moving, ever-changing environments (e.g., VPU, fanless, mini size, PoE support, extreme temperature, low power consumption, etc.). The future of in-vehicle computers must not just meet but exceed basic requirements to accurately and effectively execute constantly advancing intelligent applications.

Product Adopted:

Share to:
Comments ( 0 )
security 50