Vivotek’s AI Survival Strategy
Taking on China’s Goliaths
Taiwan’s surveillance industry companies have been battered by Chinese competition, but they are hoping to turn the tables by developing artificial intelligence systems that learn as they go and create value. Can they successfully fend off China’s threat?
Taking on China’s GoliathsBy Elaine Huang
From CommonWealth Magazine (vol. 638 )
The Wall Street Journal delivered a shock in mid-November when it reported that surveillance systems made by video surveillance solutions provider Hangzhou Hikvision Digital Technology were being used to monitor a U.S. Army base in Missouri, the U.S. embassy in Kabul and other sensitive spots. The revelation sparked security concerns, especially given that Hikvision is 42 percent owned by the Chinese government.
But Chinese surveillance products have actually been in American homes and businesses for a long time, and Hikvision, which was founded in 2001, and another big Chinese company, Dahua Technology, now have a nearly 40 percent share of the global surveillance market.
These major “red” surveillance brands have carved out bigger market shares through price wars, with the prices of surveillance cameras falling 11.4 percent annually, according to IHS Markit Technology, a leading provider of market intelligence on the global surveillance industry. Tumbling prices have decimated Taiwan’s security industry, which was once seen as an emerging star.
It has gotten so bad that one of the benchmarks of Taiwan’s surveillance industry, Yoko Technology Corp., announced at an investor conference in October 2017 that it was selling its factory and phasing out its surveillance product business.
“It used to be that Taiwanese surveillance product companies could earn three cameras for every one they sold, but those days are long gone. Every company is looking for a way to reinvent themselves,” says Owen Chen, the chairman and CEO of leading Taiwanese surveillance system supplier Vivotek Inc., in his office in the Taipei suburb of Zhonghe.
Chen studied at National Taiwan University’s Graduate Institute of Electrical Engineering and spent time in its Communications & Multimedia Laboratory (CM Lab). He was a student of Chen Liang-gee, then an NTU professor and now Taiwan’s science and technology minister.
After spending some time at Chunghwa Telecom Laboratories, Owen Chen later joined with some of his classmates from the CM Lab to found Vivotek in 2000, and it soon emerged with the strongest R&D capabilities in Taiwan’s surveillance sector.
Now, at a time when Chen Liang-gee has become one of Taiwan’s biggest promoters of artificial intelligence, his former “disciples” from the CM Lab are actually applying AI in the local surveillance and security business.
Fighting Back: Upping the Ante on R&D
Vivotek, which went public in 2011 at NT$87 a share, saw its stock price soar above NT$200 for the first time in late 2013 and early 2014. But just as the company was reaching new heights, it was dealt a huge blow by Hikvision’s price-slashing. Its operating revenue in 2014 fell 10 percent from a year earlier and its earnings per share were chopped in half.
A senior executive in Taiwan’s surveillance industry recalls that many Taiwanese companies were unable to hold off the “surging red tide” at that time and responded by downsizing their workforces. Only Vivotek’s Chen decided not only not to lay off employees but to continue to invest in R&D.
A look at Vivotek’s financial statements shows that its investment in R&D has risen every year, with R&D expenses accounting for 13 percent of the company’s revenues in 2016, higher than the 8 percent registered by TSMC, long recognized as one of Taiwan’s leading proponents of R&D.
So what opportunities has Chen identified that are worth upping the ante on research and development to fight the scourge of Chinese competition?
From ‘Seeing’ to ‘Understanding’
“Any smart home, smart factory, smart office, smart retail space – any place that is ‘smart’ – needs cameras. We believe that the market will get increasingly bigger,” Chen says.
“But the demand and uses for smart video applications are different in every vertical market, such as transportation or factory automation. So you have to customize,” Chen explains, arguing that the market will become more niche-oriented in the future.
Though that will allow companies to sidestep the cutthroat competition of standardized products and open up more opportunities, it will also require more investment in R&D, Chen believes.
Vivotek is currently targeting three vertical smart surveillance markets: retail, transportation and smart buildings.
Five years ago, even before anyone had heard of AlphaGo, Vivotek had already started investing in video analytics technology, including artificial intelligence.
At the time, Vivotek was getting inquiries from customers looking to extend the functions of standard surveillance systems. Retail store clients who bought surveillance cameras wondered, for example, how many customers were visiting their stores? They also wanted to know what were the times of day with the most customers, what entrances did they use the most, and which counters had the most traffic?
Image: Chien-Tong Wang
“In the past it was about being able to see things. Now it’s about ‘understanding what you’re seeing.’ It’s an inevitable trend. We discovered that using deep learning technology to do analysis could solve many problems, and we built a team around that,” says Steve Ma, vice president of Vivotek’s Brand Business Research & Development Division.
To capitalize on the opportunity, Vivotek set up a “Smart Video Analytics Department” three years ago that is exclusively responsible for integrating deep learning with video analytics and mining value from the large volume of video footage generated by its customers.
The company also formed a brand project team that spends a third of its time overseas doing field tests with Vivotek customers around the world.
“If customers have specific requests, we have to do on-site tests. Once they’ve agreed to the on-site tests, we take the data and use new technology to test and train the system. You need data to train it, otherwise the product won’t be any good when it’s installed,” Ma says.
Joe Wu, the company’s chief technology officer, says having a Vivotek team doing tests with customers is necessary because he has “been burned” too many times, when it appeared that a prototype’s smart video analytics seemed to work fine only to fail when the customer put it to the test.
“The floor may reflect light or a person’s backpack may lead to faulty readings. So we have to be there to test the product and make adjustments. The strength of deep learning is that the device can be constantly trained and adjusted. In the past, if a product that you thought was fine, when it shipped didn’t work when tested by the customer, it was really hard to make the necessary adjustments to get it right,” Wu says.
With deep learning training, as long as enough data exists, it is possible to train a device to meet customer requirements within acceptable time frames. And because the system becomes more accurate the more data it’s fed, customers are willing to sign off on it, according to Wu.
“The team has more than doubled in size since it was formed, and it’s still growing,” Chen chimes in, noting that many employees realize how important deep learning will be to the surveillance and security business in the future and are eager to give it a try.
Traffic Figures 98% Accurate
Last year, Vivotek launched its star product – a “3D People Counting Solution” – that caters to smart retail stores. Smart cameras are installed around the store and count the traffic at a 98 percent accuracy rate with the help of an exclusive algorithm. Customers then use the data to run their stores more efficiently, for example, scheduling shifts based on anticipated traffic or timing promotions at the most opportune times.
Vivotek’s “smart surveillance” systems are now being used overseas by retailers in Colombia and a Japanese department store chain in Southeast Asia, and at home by the National Palace Museum gift shop and others.
These strategies have paid dividends for the company, even if its stock has remained depressed below NT$100 per share since September 2014 with only a few exceptions.
As of December 2017, the company has seen annual sales growth for 31 consecutive months despite Hikvision’s ongoing price war, and its gross margin has started to rebound after a prolonged decline.
“We have entered a new era of surveillance,” says William Ku, vice president of Vivotek’s Brand Business Division. “We have shown that network cameras not only can provide security, they can also help retailers track in-store traffic through smart analytics to drive revenue growth.”
“Things have changed from when we simply sold a camera for a certain price. Now it’s about adding value. You have to identify the different needs of your vertical markets and undertake customized R&D,” adds Owen Chen.
Unlike in the past when images were transmitted from cameras to the cloud and then analyzed by an algorithm, Vivotek has also opted to use deep learning algorithms to analyze images right on the device itself, meaning that data computing and transmission can be done even more quickly, and the chance of inaccuracies resulted by having to compress data to send it to the cloud is eliminated.
“We try to complete the analysis within the camera,” Wu says.
Vivotek subsidiary and multimedia system-on-chip provider Vatics has developed a new chip with an embedded deep learning function that will be used on many Vivotek products once it hits the market next year.
The New Weapon to Fight off China
Like Vivotek, some top-flight Taiwanese startups foresee a bright future for the integration of artificial intelligence and surveillance products. They include Umbo Computer Vision, which recently raised US$6.8 million in a new series A round of investment.
“AI is still not widely used. Its penetration rate is less than 1 percent. But it is being developed at an increasingly fast pace, and about three times as many customers are seeking us out today than did a year ago,” says Umbo Computer founder Shawn Guan.
Guan got his start in the business at publicly listed surveillance system provider NUUO Inc. Over the past year, Umbo Computer’s smart surveillance products have been used by a major restaurant chain and Taiwan’s Agency of Corrections under the Ministry of Justice and helped them analyze more than 2 billion images.
“Our Light product sends an alarm when a person goes over a wall, but such things as cats or dogs going by or changes in light do not trigger false alarms,” Guan says. The system’s detection capabilities show that with deep learning, its level of accuracy is constantly on the rise.
One example: A day after Wen Huah Elementary School in Taoyuan installed the device, the school’s teachers wrote to Umbo Computer. They said they had never seen anybody scale the school’s outer wall, but after putting in the device, they noticed it happened around a dozen times in just one day.
Another smart surveillance startup – Skywatch – can be found in the headquarters of leading notebook computer contractor Inventec Corporation on Chengde Road in Taipei.
Skywatch uses artificial intelligence-driven video analytics to analyze video stored by clients in the cloud and then present the analysis to clients. The company was the first in Taiwan to commercialize video summary services in the country.
Chinese hardware providers such as Huawei and Hikvision have sought out Skywatch, hoping to incorporate its surveillance software in their devices. But beyond technological advances, size and financial resources are still necessary weapons in the battle for survival.
To keep the “red tide” at bay, surveillance sector leader Vivotek decided to take on a strong partner with deep resources, allowing Delta Electronics to buy a stake in the company that made it its biggest shareholder.
“For us, if China wants to wage a price war, we have to improve our ability to dictate pricing and our automated manufacturing capabilities,” Owen Chen says bluntly.
But an even more important reason for taking on Delta Electronics as a partner was the need for stronger backing in developing outlets for artificial intelligence and its applications, a necessity if Taiwan wants to counter the “red” supply chain.
Taiwan’s surveillance industry’s war of resistance against China’s Goliaths has floundered in recent years. But if Taiwan can brandish artificial intelligence effectively in the future, the sector may still have a chance to survive and even thrive.
Translated from the Chinese article by Luke Sabatier