Author: Li Xi
Edit: Yu Kuai
Always go to Shaoxing to experience the Jiangnan water town.
It is the most description of Shaoxing in major social media. In fact, Shaoxing not only has the tenderness of the water town, but also the gentleness of the gentlemen with the firmness. Today, it is more sturdy with technology.
This cultural capital of East Asia, which has a history of more than 2,500 years, has started a series of urban renewal action plans in recent years.
Intelligent transportation is the core of urban digitalization.
The troubles of a thousand-year-old ancient city
As the famous architect Liang Sicheng said: “The city is a science. It has meridians, pulses and textures like the human body. If you don’t treat it scientifically, it will get sick.”
The city is home to more than 40% of the population and has been sick for a long time in terms of traffic congestion.
As a prefecture-level city, Shaoxing is also troubled by this, especially the large-scale international competitions it will undertake, which have higher requirements for comprehensive urban conditions such as transportation. Under the goal of building a “three-dimensional, efficient, green, and intelligent” modern comprehensive transportation system, Shaoxing chose to join hands with SenseTime.
Why did Shangtang choose Shaoxing?
Third-tier cities with more than 5 million permanent residents and hundreds of intersections on the road network scale.
Shaoxing’s urban sketches are extremely typical, with the largest number of cities of this size in the country, reaching several hundred.
Shaoxing traffic intersection topology map
In terms of traffic road network conditions, compared with Beijing, Shanghai, Guangzhou and Shenzhen and the provincial capital cities of tens of millions, there are relatively few social factors and it is easy to transform and upgrade. At the same time, the density of traffic cameras in Shaoxing is moderate.
Shaoxing is equivalent to a large-scale traffic sample, which can be replicated across the country after successful implementation.
But upgrading the millennium ancient city is not an easy task.
First of all, it has always been a situation that the transportation field has faced for a long time in the past, focusing on construction and ignoring application.
Construction first, software last. The first is to ensure that the business of the system is feasible, and the second is the efficiency of the system, and there is even no “secondary”.
Therefore, there is a lack of systematic planning and deployment, a lack of systematic training of transportation software system management talents, and a lack of equipment access network procedures and systems. This makes the development of informatization and transportation soft power out of touch.
Second, the traffic perception ability is insufficient.
At present, the city is mainly based on the use of the old, which means that after analyzing the existing video, the results are closed with the signal machine, and the deployment density and location of the cameras are relatively high.
Most cities, including Shaoxing, have several problems. First, in the early stage of urban construction, transportation projects are deployed in multiple phases, and the camera models and video resolutions are different. The second is the location and density, and the density of camera placement is uneven.
Some intersections only have cameras in one direction, lacking the ability to convert video into more and richer traffic data to support traffic analysis, police situation discovery and traffic control.
Furthermore, the coverage rate of Shaoxing cameras and the system networking rate of traffic lights are relatively high, but the degree of traffic intelligence is low. The camera is only used for illegal capture, and cannot provide traffic flow and other data information required for intelligent traffic operation management. In other words, the traffic department can’t even count how many vehicles are on the road in Shaoxing every day.
Due to the lack of perceptual data support, most traffic signal solutions rely on manpower and expert experience.
At each intersection, it is necessary to manually investigate the time period of traffic flow at the intersection in advance, and then formulate a relatively reliable plan for different time periods, and finally manually implant the signal control system. Taking first- and second-tier cities as an example, 300-500 people need to be deployed.
SenseTime also found in the survey that in many cities, after the installation of signal lights, the manufacturer will initialize the timing parameters for each intersection. The traffic demand has been constantly changing over the years, but the intersection signal light scheme, some rely on manual settings, and some even It stays in the stage of initialization parameters of the manufacturer and has never been adjusted.
“It may not change for a week, a month or even a year, and 80-90% of the cities in the country are in this state.” said Guo Haifeng, general manager of SenseTime’s intelligent transportation product line.
In this way, not only the time-scheduled plan for each intersection is solidified around the clock, but also the basic traffic signal plan is not refined enough. The average number of timing plans for each intersection in third- and fourth-tier cities is less than 4 sets.
In addition, junction information (canalization, traffic facilities, signal light settings, etc.) lacks ledger management, and is recorded in manual tabular form at most. If it is not updated in a timely manner, management is difficult to sustain, resulting in the inability to effectively describe the invisible assets of the intersection.
All these have led to the development of intelligent transportation for decades, and it is still unable to output reliable traffic flow at intersections, and it is difficult to implement perfect traffic control theory in the real traffic environment.
In fact, these problems are not unique to Shaoxing. Shaoxing and other cities have solved the problems from scratch in the construction of transportation informatization in the past few decades, but the problem from existence to excellence has always been in its infancy.
Come early is worse than coincidence
The last time traffic received the keen attention of the outside world was around 2016, when the big Internet companies came to the forefront and entered the traffic field.
At that time, they all relied on their own map software data, but due to the large difference in the coverage of vehicles on the road network of different cities in the C-side data, the data sample size was insufficient.
In addition, different systems and departments are difficult to get through. The use of this data by the transportation department involves switching between the external network and the internal network, which not only has the problem of data delay, but also hidden dangers of data security and privacy protection.
Therefore, it is difficult for the C-side Internet data model to comprehensively evaluate the traffic situation, and only a macro-level timing plan can be obtained, which cannot form a closed loop with the signal machine and signal system, and the landing effect is far lower than expected.
In the final analysis, this wave of transportation upgrades is only information-based, not intelligent, and there is no large-scale application of AI technology.
After the explosion of AI 6 years ago, it still cannot support the transportation intelligence giant. Since 2018, AI has finally made breakthroughs after going through a big pit, and the technological maturity and usability have reached an appropriate node.
During this period, SenseTime has also accumulated a lot of experience in AI technology reserves.
It is better to come early than coincidentally. Although SenseTime entered the transportation field in 2019, it was just the right time.
Transportation is a very special market in which AI is flooding into thousands of industries: as much power as possible, so much resistance.
It stands to reason that with policy support, scenario potential, and upgrade needs, the basic plan is correct, but in the past 20 years, not only have there been no giants, but also no absolute leading companies.
The intelligent transportation system is extremely complex and involves multiple dimensions. Here we will discuss two issues.
The technical dimension, the technical peak of decision-making intelligence.
The full perception of urban computing consists of perceptual intelligence, cognitive intelligence, and decision-making intelligence.
At present, the most widely used AI, such as image recognition, language recognition and computer vision recognition, are perceptual types. The core logic is input and output functions. The logic is relatively simple, and there are still many unsolved problems.
Applications such as intelligent transportation and AI game competition are decision-making. The core is to constantly re-judgment and reason based on a lot of knowledge, and to make decisions, which are more dynamic, flexible, and more complex.
To close the loop in a transportation system, AI may need to consider the endless problems that come with the combinatorial explosion:
How to coordinate a signal light at an intersection?
How to determine the scope of synergy, is it a block, a district or an entire city?
How to find the cause in time, evacuate in time and prevent it afterwards?
Where is the root cause of long-term traffic congestion and how to solve it?
How to formulate urban transportation plans in a short period of time when large-scale events are held in the city?
The above problems require more sophisticated maps, physical perception, and decision-making AI with strong task intelligence to solve them.
In the business dimension, the traffic demand is huge, the degree of fragmentation is high, and the marginal cost is high.
Industry experts told AI Nuggets that there will be more than 100,000 algorithms in smart cities in the future, and transportation is a big demand. It is reported that there are currently more than 100 algorithms in the transportation field.
Shang Tang wants to climb this technical peak.
In 2019, SenseTime deployed perceptual intelligence. This year, it further opened up the gap between perceptual intelligence, cognition and decision-making intelligence, and added fire to the technological explosion of decision-making intelligence.
This business gully, Shang Tang also wants to fill it.
There is a huge amount of scene data in the transportation field. It is fragmented, the project scale is small, and the revenue is small, but the quantity is large. Customization and low delivery efficiency have become the pain points of the industry. To solve the long-tail scene, the fragmentation problem is solved.
The essence of fragmentation lies in the low versatility of AI models, “hand-workshop” model production, cumbersome AI project development processes, long data and training time, high model customization, low reuse rate, and repeated wheel creation in different scenarios.
This is the task of the underlying infrastructure.
SenseTime’s AI large-scale installation attempts to make the AI model realize the assembly line production mode and reduce repeated research and development.
In algorithm production, while accumulating a large number of models, precipitation of reusable algorithm underlying modules, and modular production, while generalizing the large model + small model mode, it reduces customization, improves AI versatility, and reduces the marginal cost of entering new scenarios.
In the R&D process, AutoML automatically performs some engineering tasks to lower the threshold for machine learning and reduce the need for AI experts.
From the bottom layer, the algorithm models of various scenarios are abstracted, and various algorithm tool modules are used to customize new scenarios in a modular way by combining algorithm suites, so as to produce and deploy AI models on a large scale at a low marginal cost.
The pulse of the city beats rhythmically
Let’s take Xinkong as an example to go deep into the details of transportation to see the intelligent upgrade of Shaoxing transportation.
In Guo Haifeng’s view, the essence of traffic lights is that the first policeman maintains the order of the intersection, and it solves the problem of order, not efficiency.
Under the premise of prominent traffic contradictions and limited space resources, allowing the road network to carry more traffic flow is one of the cores of transportation. At the same time, under the optimal state of system operation, it can ensure management needs, travel needs, and driving experience.
The previous products were based on equipment and produced “material products”, which freed manpower. In the future, products will be created based on data, and “intellectual products” must be produced to liberate brain power.
“Like in the era of smart phones, the upgrade of functional phones to smart phones is not only an upgrade of software systems, but also an upgrade of the entire industry such as mobile phones.” Guo Haifeng said by analogy.
This is a question that SenseTime has been thinking about since it entered the transportation field, and found the core entrance of “signal control”.
“To a certain extent, it can be considered that information control is the core of traffic. Under the condition that the basic conditions of the road surface and the traffic regulations remain unchanged, the information control is the only variable that can increase the upper limit.” An expert once told AI Nuggets.
According to SenseTime’s internal simulation and actual measurement in some cities, the performance of the signal control system can be improved by 10%-20%. A 10% increase can theoretically reduce the one-hour commute time to 54 minutes, which will affect the labor value and economic value generated by millions to tens of millions of people in a city.
In other words, if the traffic network is the capillaries of a city, the traffic lights flashing 24 hours a day and night are like the clock and pulse of a city. The flow of people, vehicles, and logistics throughout the city follows the pulse.
The urban traffic in this mode is no longer a passive response management thinking, but a systematic operation thinking.
SenseTime has three standards in the transportation digital operation system: benefiting from the old, reducing costs, and improving efficiency.
On the front end, SenseTime has two actions. One is to benefit from the old. Based on the leading AI analysis capabilities, it can identify multi-dimensional traffic incidents, traffic violations, analyze traffic parameters, and revitalize the stock video. The second is to reduce costs and liberate manpower. From perception AI to recognition Know AI. The former relies on perception AI capabilities to help users build the most complete traffic operation data collection support system with the lowest investment, while the latter uses AI to replace a large number of human resources in the whole process of signal optimization, realizing the upgrade of signal optimization services from labor-intensive to intelligent .
At the back end, based on the digitization of the urban road network, combined with the dynamic data of traffic operation data, the networked control of the city’s global traffic is completed, and the urban operation efficiency is truly improved.
To achieve this goal, Shangtang has several magic weapons.
1 AI platform, SenseFoundry Tran Shangtang Ark Traffic Open Platform provides platform-level solutions for intelligent monitoring in the transportation industry, builds a base for intelligent analysis of traffic views and big data analysis of views, and empowers urban traffic management and control services from a multi-layered perspective.
A traffic digital base realizes the digital management of dynamic data such as urban traffic static road network equipment, facilities and resources, traffic parameters, traffic events, information release, and traffic status. The 225 signal intersections in the main urban area of Shaoxing were able to form digital intersection files, which greatly improved the efficiency of equipment operation and maintenance.
For the segmented scenarios in the transportation field, SenseTime meets them one by one through the N+ business system.
Facing traffic safety management, SenseTime’s Ruitu series products provide the basis for traffic early warning and refined law enforcement, and at the same time predict traffic safety risks. Through real-time detection and big data analysis, active early warning of traffic accidents and hidden traffic incidents, real-time linkage of traffic management and disposal, enhances the ability to actively detect and quickly deal with police situations.
To ease traffic jams and ensure smooth traffic, SenseTime believes that it is necessary to ensure that it has excellent tuning tools first. Its RuiControl series products focus on traffic signal tuning, and formulate control strategies for traffic data conditions in traffic scenarios. Layered and hierarchical control strategies and The control scheme, in terms of actual operation, is divided into three categories: AI automatic control, AI auxiliary control and basic control. The degree of manual intervention is selected according to different intersection conditions.
Secondly, it is necessary to ensure that there is a reason to be found, and SenseTime think tank products combine AI technology, machine vision technology, data mining technology, big data analysis technology, traffic consulting case library and traffic expert experience to realize AI consulting instead of traditional manual consulting. The liberation of traffic manpower, through the utilization of traffic data perception, the establishment of multi-dimensional traffic system, and the establishment of traffic management knowledge map, realizes problem detection in seconds, comprehensive diagnosis of congestion, system determination method and long-term evaluation effect.
After deploying SenseTime’s large-scale video analysis system, Shaoxing not only obtained the 24H traffic operation status of 109 signal-controlled intersections, but also the flow, traffic capacity, and saturation of each turn at each entrance. These are the details of the intersection. the premise of the analysis.
After high-quality traffic data information is in place, SenseTime’s traffic base and Shaoxing information control platform, in addition to real-time perception, correlation of traffic events and communication changes, and timely warning, help traffic police make intelligent traffic decisions and control.
In this way, 0 distance between video and signal control is realized.
Legend: Smart recommendation for solution optimization (cycle and phase duration in seconds)
In the end, SenseTime established a holographic traffic digital base with the lowest investment on the basis of Shaoxing’s digital road network, and completed a breakthrough in Shaoxing’s large-scale traffic parameter analysis from 0 to 1.
Legend: Significant increase in travel speed
Shaoxing is just a microcosm of hundreds of cities in China. In Guo Haifeng’s view, a good industrial-grade software product should meet four conditions: security, stability, robustness, and user experience.
In the field of transportation, safety is always the first priority; information control systems, especially large-scale engineering software, have very high requirements for stability; and because data quality is uneven, phased deletions and other data anomalies will always exist, which requires The algorithm has high robustness, otherwise the generated scheme will deviate, and the implementation of the signal control system will be difficult to withstand a long-term test; and the user experience requires flexible system operation.
This is also the basic requirement of SenseTime’s digital products. At the same time, SenseTime’s entire set of digital operation product matrices are interconnected to systematically improve the efficiency of transportation operations.
The Transportation World Needs Smart Operating Systems
Strategizing in the tent, the decisive victory is thousands of miles away.
For thousands of years, people in the world have imagined that Zhuge Liang would sit quietly in the tent, shake his feather fan lightly, calm down, talk about the direction of the war thousands of miles away, and decide the rise and fall of the world, whether it is happy or not.
But is this really the case? We have ignored many details in history intentionally or unintentionally. Behind the “strategizing” is the result of many comprehensive factors.
Taking history as a mirror, in the wonderful story, Zhuge Liang can calmly analyze and deal with strategies in the account, which is nothing more than intelligence, strategy and talents.
To truly improve the efficiency of the transportation system, it also requires the real-time aggregation of massive data, AI capabilities, and platform capabilities.
In the past, the essence of transportation digital construction was IT construction and lacked operational thinking, which is why the dividends of technology did not seem to bring qualitative improvement to transportation.
Behind the operating system, what is really lacking in the transportation field is an “intelligent operating system” and a “professional transportation software provider”, which can truly wake up equipment and reduce costs and increase efficiency based on actual traffic conditions.
This is what SenseTime is doing and will always insist on.
Under this theory, SenseTime intelligent transportation has penetrated into the urban fabric.
In the subway field, Shangtang’s non-inductive pass products for entering and leaving stations have been implemented on more than 30 lines across the country; in the high-speed rail industry, Shangtang’s AI safety inspection products have analyzed a total of 50+ lines with a total of about 20,000 kilometers; in the field of traffic management, Shangtang’s traffic management Jizhi products have also landed in more than 20 cities, among which the traffic AI platform has access to nearly 10,000 videos; in the high-speed field, SenseTime AI audit products have deployed 30+ toll stations.
Although there are many difficulties in intelligent transportation, it does not affect its high expectations because of its amazing size.
According to the forecast of the Prospective Industry Research Institute, in the next five years, China’s intelligent transportation market will continue to maintain a rapid growth trend.
The market size of the intelligent transportation industry in 2010 was only 20.92 billion yuan, and it reached 51.59 billion yuan in 2017. It is expected that the market size will reach 145 billion yuan in 2023.
In 2019, the Central Committee of the Communist Party of China and the State Council issued the “Outline for Building a Strong Transportation Country” to vigorously promote intelligent transportation. China’s new infrastructure, dual-carbon strategy, and transportation construction will be one of the primary construction tasks.
Through the ages, all industries have prospered, and the first is in traffic. It is already indisputable that transportation will be the star of the future.
We will wait and see how AI companies such as SenseTime will promote the digital upgrade of transportation. Leifeng Network Leifeng Network Leifeng Network
This article is reproduced from: https://www.leiphone.com/category/smarttransportation/uoLgcL8v9nU5Cvez.html
This site is for inclusion only, and the copyright belongs to the original author.