AI driver monitoring is pushing forward semi-autonomous vehicle adoption
Jun 5, 2019
In October 2018, a fight erupted between a passenger and the driver of a Chongqing bus. In the hubbub, the vehicle swerved across the road and through the barrier of a bridge, falling into the Yangtze River. All 15 people on board perished.
The incident sparked widespread indignation across the country as observers largely blamed the passenger for provoking the fight. It also brought to light some 20 other attacks on bus drivers that occurred in China that year, although none with so high a casualty count.
Against the backdrop of the public outcry, officials took action. “[The] government regulations’ requirements are higher now,” Liang Kun, product manager at Xiamen-based surveillance and security firm Reconova, told TechNode. Passenger aggression towards drivers can now be punished by law, and the installation of “active safety” technology is required on commercial vehicles in addition to public transportation. Some of the new measures could prevent tragedies like the one that happened in Chongqing, Liang believes.
Along with the heightening of regulations surrounding road safety, Reconova has seen more demand for its driver surveillance services, which include facial recognition devices that detect distracted driving as well as machine vision technology that surveys and warns of nearby vehicles and pedestrians.
The six-year-old startup, which completed a Series B last May led by Intel Capital, is part of a larger trend towards machine-assisted driving. As applications for fully-autonomous vehicle technology–for consumers, at least–proceed relatively slowly, this particular sector is accelerating, with a growing number of players, including artificial intelligence (AI) giants Sensetime and Baidu, entering the market. As a result, increased surveillance of drivers could significantly reduce the likelihood of accidents; however, it also raises certain security risks as well as potential concerns over privacy.
Eyes on road safety
On a sunny afternoon in Shenzhen, Guangdong province, TechNode joined Liang for a spin in a Reconova test vehicle. Inside five cameras were plastered in a straight line down the van’s windshield, each one able to detect movements by nearby vehicles. Another device above and to the left of the steering wheel was pointed directly at the driver’s face, checking for signs of drowsiness, phone use, or smoking.
At periodic points during the drive, Liang demonstrated how the system reacts to various risky behaviors. Twice, after checking that the road is clear, he closed his eyes for a few nerve-wracking seconds before the system’s speaker barked a reprimand: “Danger, please be careful.”
Holding a phone to the side of one’s face while the van is moving elicits a similar warning, as do too-quick turns and neighboring vehicles that switch lanes without leaving enough space. In the relatively calm mid-afternoon traffic, though, the system is mostly quiet, only occasionally blaring out brief cautions.
According to Liang, camera footage of driver misdemeanors and other safety risks can be automatically uploaded to a company’s platform if the system is online.
“In accordance with Chinese law, the equipment doesn’t collect the personal information of the driver or the person being surveilled,” Liang said in reference to Reconova’s facial recognition technology, which can also verify drivers’ identities.
“We don’t know who is who,” he added. According to him, the system doesn’t cross-check images with ID information, but only checks whether someone’s facial characteristics match companies’ driver records.
This year, a major Chinese logistics company secured Reconova’s services for a part of their delivery fleet. “Our first batch has already been installed and their testing program was excellent,” Liang told TechNode. “If it really is effective,” the client has plans to expand, he said.
The company has also had “successful use cases” in the area of public transportation. Bus company clients, for instance, can install a one-click panic button on their vehicles, allowing drivers to contact police more easily in case of an emergency. Another, optional feature allows buses to be brought to a halt via remote control.
Reconova sales director Morgan Guo told TechNode in an interview that this field has grown rapidly in the last year: from 4,000 orders in 2017, demand soared to 30,000 devices installed the next year. In 2019, Guo predicts, that number could grow another “70-80%.”
In addition to general public safety, increased scrutiny of truck and bus drivers is also good news for companies like Reconova, transportation firms, and insurers. The reaction of the drivers themselves, however, has been mixed.
“Drivers will use things to block this device, or bend the device around so that it’s not effective,” Liang told TechNode while gesturing to the facial recognition gadget to his left. Because employees feel that “there’s something monitoring their behavior,” Liang says, “there will be aversion.”
Hiko Lee, enterprise solution manager of GreenSafety, a startup that supplies similar driver surveillance systems to business clients in Hong Kong, Macau, and Taiwan, has heard of similar resistance from drivers.
For clients such as electricity supplier China Light and Power Company (CLP), GreenSafety assigned drivers in 50 vehicles grades based on their behavior.
“When the score is high, around 100 marks, then the performance is good” while 50-60 might be the mark of a “bad driver,” Lee told TechNode. Thanks to improvement in driver ratings over time, GreenSafety won the chance to trial their devices for two major bus companies in Hong Kong. Currently, its systems operate on around 400 vehicles in the city.
“Of course at first they really don’t appreciate it,” said Lee of CLP’s drivers. After three to six months of education, however, attitudes slowly changed.
“The Hong Kong bus and taxi drivers may work over 10 hours per day. So we will teach them by training, by lessons, by different methods–maybe talk to the management and help the management to persuade them,” Lee said.
He compares the situation to the widespread adoption of GPS tracking and basic in-vehicle cameras over the last decade. Drivers gradually accepted the initially intrusive technology because “they know that this kind of system can protect them” from liability in accidents.
Five months before the Chongqing bus fell into the Yangtze River, killing 15, another case of driver-passenger violence attracted national attention. In May 2018, a woman using online ride-hailing platform Didi to hitch a ride was murdered by her male driver. Just a few months later, in August, another female passenger using the same service was raped and murdered by the man behind the wheel.
The incidents sparked a nationwide backlash against Didi, and provoked official scrutiny—leading the platform to adopt a series of new safety measures, from an emergency number linkup for passengers to optional video or audio recording of rides.
Asked whether high-tech AI features might soon enter ride-hailing companies’ arsenals, both Liang and Lee said the industry showed potential.
Companies in the field are currently in talks with Reconova over facial recognition solutions to verify drivers’ identities, according to Liang. In both of last year’s high-profile Didi murders, the culprits posed as registered drivers on the app. “This need exists,” said Liang.
Tal Krzypow, vice president of product management at Israeli computer vision firm Eyesight, says China’s ride-hailing market is just as interested in driver surveillance as “any other fleet.”
Eyesight is currently working with original equipment manufacturers and aftermarket partners to provide driving monitoring system solutions to China. “There is a willingness to adopt new technology and going to market quickly is very impressive” in the country, Kryzpow told TechNode.
Using advanced and often expensive technology such as machine learning to analyze video footage, however, may not be on the table for those companies as of yet. Lee pointed out that ride-hailing startups may not be inclined to invest so much in individual cars and drivers. However, with pressure from government as well as popular sentiment, that could change, Liang said.
Lee also foresees a larger shift to the consumer market as driver surveillance technology continues to advance. Once more affordable, accessible devices are released on the market, “maybe the customer can just buy it from the Internet and they can install it themselves very easily.”
In a written statement compiled for TechNode, analysts from international firm BIS Research predicted rapid growth of connected and partially autonomous vehicles in China over the next two years. As a reference, they cited the Chinese government’s prediction that the domestic market for connected auto will grow to $14 billion by 2020.
However, the increasing amount of data will also require cybersecurity upgrades. “Vehicles need protection from threats such as malicious software, unauthorized access, attack on vehicle CAN [controller area network] BUS and ECUs [electronic control units], sniffing of vehicle data, loss of cloud data, and malicious codes in the vehicle, among others,” BIS analysts wrote.
Speaking of another sector of Reconova’s, smart security and surveillance systems for corporate and official clients, Morgan Guo said that “privacy will be protected.” According to Guo, the company itself doesn’t permanently store visual or other information gathered by its software, although he admitted that China’s government is by law allowed to do so.
Currently, more than 200 electric vehicle manufacturers, including Tesla, BMW, Volkswagen, and Nio have been called upon to transmit their vehicles’ location data to government-backed monitoring facilities.
That raises the question of where the data gathered by systems like Reconova’s and GreenSafety’s will be stored, and who will have access to such valuable information. Generally, how the technology is implemented is left up to buyers.
“We do provide guidelines,” said EyeSight’s Krzypow.
As Berkeley professor Alexandre M. Bayen, who directs the university’s Institute of Transportation Studies, told TechNode, however, individual drivers’ data privacy could already be compromised. According to Bayen, the issue “in a sense started 10 years ago.”
He referred to the advent of smartphones, as well as the data-gathering that accompanies their use: “Your phone activity while you’re driving, potentially the onboard car activity if your car is somehow hooked up with your phone to Bluetooth or any other link.” “All that data, it’s already there, it’s already available,” and being accessed by large tech corporations like Google, Bayen added.
“With more data, of course, the problem grows,” Bayen said. But he believes that the ultimate responsibility of protecting that information falls on the government. “To me, the technology is just a means to reveal the data; the real question is the question of policy,” Bayen said.
With additional reporting by Chris Udemans.