XMOS releases low-cost, low-power automatic license plate recognition (ALPR) reference design for smart parking

British chip company XMOS recently announced the launch of its Automatic Licence Plate Recognition (ALPR) reference solution, which aims to drive ALPR in parking lots from complex, resource-intensive hardware to simple, device-based artificial intelligence (AI) solutions, making it extremely easy for equipment manufacturers and system integrators to produce, install and integrate automatic license plate recognition systems.

This reference design was developed in collaboration with the algorithm team of Tongji University and a local Chinese solution company, and can achieve high-precision reading of slow-moving license plates within a distance of 3-5 meters. Thanks to the powerful AI capabilities of XMOS's xcore.ai chip, lightweight machine learning models can seamlessly adapt to work in low-power, low-cost scenarios without compromising accuracy.

Traditionally, the hardware integrated in parking lots using ALPR functions far exceeds the specifications required for low-speed, close-range license plate recognition. High-resolution cameras and complex machine learning models run on powerful processors, and these models often rely on the background cloud for image processing, which makes the implementation cost of ALPR extremely high in many cases. Comprehensive innovation from chips to devices to systems is required to reduce procurement costs and usage costs.

相比传统的ALPR解决方案，XMOS的参考设计可以在设备端提供边缘计算所需的精度和算力，从而显著降低功耗和减少物料清单（BOM）。通过消除对高成本硬件的需求且几乎消除对云连接的需求，这种设备成为了整个智慧城市ALPR基础设施中切实可行的组件，可便捷地部署在各类智能停车场所，或者嵌入到智慧园区和智慧城市等应用中，同时使用成本大幅度降低。

“Cloud connectivity and massive processing power are completely overkill for smart parking,” said Aneet Chopra, vice president of products, marketing and business development at XMOS. “This makes ALPR networks far more expensive than they need to be, makes maintenance more complex, and is fraught with privacy issues inherent in the cloud.”

“The reference design we developed eliminates these issues by simplifying the process. If you can do the required accuracy and computation on the device, you can avoid sending all the raw data to the cloud or using hardware that is too expensive or powerful. In the long run, this will help us advance ALPR in the smart parking market.”

"Simplicity and affordability are two priorities in the ALPR field, not only to drive sales but also to encourage innovation," said Professor Shaoming Zhang of Tongji University. "Making devices cheaper, simpler, and more reliable is important for smart cities, while reducing the size of machine learning models so that they can run on chips like xcore.ai that can be mass-produced will provide developers with funding and design flexibility to experiment."