AI Emerging Company

Seemmo Sees More: Identifies 28,000 Distinct Vehicle Types on Chinese Roads

At the recent youth and technology themed 2050 Conference in Hangzhou, Seemmo CTO Jianhui Wang delivered a talk called The Survival Guide for AI Companies.

At the recent youth and technology themed 2050 Conference in Hangzhou, Seemmo CTO Jianhui Wang delivered a talk called The Survival Guide for AI Companies. Wang is in a position to know — his AI and computer vision company produces vehicle recognition tech that is being leveraged by the Chinese government in a wide range of public security applications. Seemmo is also active with various transportation agencies, and the company’s solutions are deployed at community centres, hospitals, and shopping malls.

There are 30,000 tech companies in China’s thriving public security market, led by state-owned behemoth Hikvision, which posted CNY41.9 billion (US$6.3B) in revenue last year. IHS Markit reports the country has more than 176 million surveillance cameras producing 66-terabytes of data every second. Some 20 million cameras are owned by the Ministry of Public Security.

Seemmo began developing algorithms for its vehicle recognition technology in 2012, and software products soon followed. In 2015, the company added pedestrian and object detection functions alongside vehicle recognition, creating a ”Video Structure Platform” for public security. Its R&D team has built low-power and high-performance hardware products suitable for edge or cloud computing.

Seemmo’s AI-powered software detects pedestrians and vehicles from raw video and image data taken from gas stations, parking lots, in-vehicle cameras, handheld devices and surveillance cameras. The system can also identify the countenances of vehicle drivers and passengers. It adds secondary attributes to identified targets. After extracting textual information, the system can perform search, data mining, and event forecasting functions. Attributes include clothing colour and style, beard, glasses, wearing a hat, carrying an umbrella, etc. for pedestrians; and brand, model, year, colour, and license plate number etc. for motor vehicles.

The company’s vehicle recognition system distinguishes itself by identifying minute details and differences, paying attention for example to inspection stickers, car decorations, luggage racks, collision dents, spare tires, and even tissues on the dashboard.

image (18).png
Identifying distinct differences in vehicles. Image retrieved via Chinese Academy of Sciences Institute of Automation

Wang told the 2050 Conference audience that there are 28,000 distinct vehicle types now driving on Chinese roads. Some, such as the 2014 and 2015 model BMW 520, look exactly the same. Wang’s team has counted 5,580 variations on vehicle fronts and 3,500 on vehicle rears over the past six years, while tracking some 400 new additions per year.

The Seemmo system can be adapted to or integrated with a variety of hardware, such as Hisilicon 359A, the NVIDIA TS 2, and the Bitmain BM1680.

This February the Shanghai Traffic Management Bureau held a license plate recognition contest with a dataset of 20 million images. In 46 hours, the Seemmo team trained their model to an accuracy rate of 99.3 percent in daylight and 98.7 percent at night.

Seemmo AI’s license plate scanning in the central city of Xi’an helped authorities catch six taxis sharing the same license plate numbers — a scam that had been very difficult for policemen to detect. Wang explained the system can also aid criminal investigations by quickly identifying criminals’ escape routes and spotting possible accomplices based on their vehicles’ driving behavior. If for example the routes of multiple drivers/vehicles with criminal histories converge, the system can issue a warning regarding a possible premeditated crime.

Wang concedes that as algorithm and video decoding complexities increase during the process of optimization, the entire system will inevitably meet a performance bottleneck. He believes Seemmo must therefore also contribute to front-end development and local area network (LAN) planning in order to help development in the public security sector.

Says Wang, “Our company mainly does artificial intelligence technology plus computer vision. The three ‘eyes’ in the [Seemmo character] ‘瞐’ actually reflect the third eye. We want to be the third eye to protect humanity. Security is the third eye of human beings, it will protect people.”

The number of security cameras in China is expected to reach 626 million by 2020 as they continue to play an increasing role in the country’s crime deterrence and evidence collection strategies. Although today’s technologies are not mature enough to provide reliable early warning or real-time crime analysis across a country as populous as China, Wang says soaring market demand will create many new opportunities in public security, and Seemmo is ready with a game plan.

Source: Synced China

Localization: Tingting Cao | Editor: Michael Sarazen

0 comments on “Seemmo Sees More: Identifies 28,000 Distinct Vehicle Types on Chinese Roads

Leave a Reply

Your email address will not be published. Required fields are marked *

%d bloggers like this: