Self-Training Telemetering Camera Module(AIY Telemetering)

The AIY telemetering module provides a low-cost, low-power, and small-sized offline image recognition module. This module compresses original parameters of deep learning models by means of the model compression for deep neutral network. Neural network can thus run on existing low-performance and low-cost chips. Our module can be easily installed in smart meter products that are heavily influenced by price and power consumption, such as pointer meter, digital meter and scale meters, etc. In this way, the module enables companies to use image recognition technology at a lower cost and to form a wireless IoT node. The core competitive advantage of this product lies in its low price and low power consumption. Besides, it offers more practical and cost-effective meter identification and data collection technology for artificial intelligence products whose market is strongly affected by price and power consumption. In brief, this module achieves low power consumption by greatly reducing the computing unit mainly on the strength of deep learning compression technology. At the same time, it can run on the existing low-cost and low-performance modules with the help of compressed model, allowing a significant reduction in costs. With power-saving and low-cost features, this product can be combined with Zigbee, Zwave, BT5.0, LoRa, WiFi, Ethernet, RS232/485/422 and other wireless and wired networks. Our product proffers a perfect solution for upgrading traditional meters to smart meter products.

Product Description：
（1）Embedded image recognition module
Our self-developed image recognition algorithm engine based on deep learning enables the algorithm to run locally on the chip without using the Internet and rapidly achieves localized data processing. The highly-integrated module is not only low in cost and power consumption but also convenient for customized development. The hardware accords with standard modular design and reserves I/O, UART, SPI and other expandable resources for tailored service.
（2）Cloud training for AI image recognition
Image recognition technology is an important field of artificial intelligence. We innovate a brand-new cloud training platform based on AI image recognition for more complicated scenarios. Our product extracts and recognizes various targets and objects in different modes quickly and accurately. In our product, there are standard SDK development kits, open data interfaces, and accesses compatible with ordinary cameras. Data processing is carried out on our cloud platform and the result is returned to the device terminal. Our product possesses the advantages of powerful extendibility, high efficiency, low cost, and strong compatibility.

Application scenarios
Products are widely used in the fields of instrument recognition and intelligent control in factories, agriculture, medical treatment, chemical industry, metallurgy, animal husbandry, fishery, forestry, households, enterprises and smart cities.
Model Cases
Industrial meter reading system of the Internet of Things
Based on the embedded image recognition module, the industrial meter reading system of the IoT automatically shoot the display panel of the meter. Relevant digital information will be extracted and recognized according to the image taken, and will be output through the standard UART port. Our product provides a fast and convenient way to connect to the Internet of Things, which allows remote meter reading and monitoring.

Advantages:
1. Non-contact mode, no need to modify the original meter equipment.
2. Stable and reliable recognition algorithm is operated locally without the need for internet connection.
3. High compatibility, supporting information-based remote meter reading of instruments from different manufacturers.
4. The recognition accuracy of a single picture is as high as 99% (enclosed environment with no external disturbance).

Technical specifications
1.Image input interface: MIPI，CVBS，HDMI，RGB
2.Input/output interface: SPI*1，I2C*2，USB2.0*1，RS232*2，GPIO*4
3.Current consumption:
when operating, 5V@300MA (DC power supply, battery or solar panel)
when waiting, 5V@0.1MA
4.Dimensions: without camera, 60mm*85mm (TBD)
5.Training environment requirements: a high-performance Windows computer with USB interface