The universal code of smart security, the three "threads" in Huawei's hands

There is an online community in the United States that enjoys untangling tangled yarn balls. Members from all over the world even buy tangled yarn balls online and then try their best to make the tangled yarn orderly. Of course, they can't use scissors.

It sounds a bit nonsensical, but in reality, people always intentionally or unintentionally put themselves in all kinds of "trouble" with tangled threads.

Take the smart cameras that everyone is familiar with, for example. With the continuous increase and iteration of various algorithm models, solutions, software and hardware products, industry managers who are full of expectations for AI are also facing increasingly complex choices and can easily fall into a state of panic and confusion.

Similar cases are everywhere. During the epidemic, most public areas and industrial parks introduced intelligent recognition systems to detect body temperatures and improve travel efficiency. But at the same time, many places have found that the deployed cameras cannot collect the data needed for decision-making, or they are over-deployed in areas such as public trash cans, and residents have a poor experience and turn around to complain... These "tangled wires" in AI applications are like the wires entangled in the sky in the old alleys. Various contradictions and situations are intertwined, making the implementation of AI confusing and difficult, and may even become useless.

As the industry gradually deepens, there are already many solutions to these "tangled threads". In the final analysis, it is to find the "thread" that can solve the contradiction, sort out the rules and order behind the implementation of AI, and then resolve the complex problems in reality one by one.

Huawei Good Hope Autumn Tasting Conference shared a lot of practitioners' practices and thoughts through real cases of AI implementation, and delivered the key "threads" that can continuously open the "tangled ball of threads" to the industry.

The first thread: the full-stack capabilities that the industry calls for

Imagine, when you get a ball of thread that is "inseparable and tangled", how would you untie it? Blindly starting to untie it will only make the thread more tangled. Normally, the first step is to unfold it a little more fluffy, and find the thread that can solve the problem among the tangled threads.

Similarly, in the process of AI implementation, it is very important to keep the entire work within a range of options with sufficient expansion space. However, in reality, AI projects always run into obstacles everywhere.

Take university cafeterias for example. After the start of the semester, many schools have seen takeaway delivery boys and students engaging in “underground transactions” across the fence, and cafeteria security guards trying to stop them, which has added a lot of trouble to the daily management of the campus. The reason is that the conventional management model has overturned in the face of prevention and control measures.

The school needs to control the number of people dining in each cafeteria, but during peak meal times, students do not know how many people are in the cafeteria, and the cafeteria does not know how many meals to provide. As a result, students often rush over after class to find that there is no food, and they can only rush to the next "battlefield".

The school also considered building a smart cafeteria to improve service levels, but after a round of research, it was found that it was not because of hardware limitations. Traditional vertical binocular cameras installed at the door are easily blocked by air conditioners and other equipment, and glass doors also increase installation risks; or the algorithm is not up to standard. The algorithms of some cameras will count people wandering at the cafeteria door as traffic, and cannot provide decision-making support for meal preparation. Some are simply "too expensive". The new campus covers an area of ​​100 acres, and there are more than a dozen student and faculty cafeterias, and there are many points that need to be deployed. If expensive cameras are used, the cost of transformation will be too high, and the data scale, storage, and computing power requirements will also rise sharply.

These complex factors are intertwined to form an increasingly tangled "ball of thread" that traps AI projects.

The key to solving the problem is to find an "all-round" support that can provide diversified choices in hardware, algorithms, solutions, etc.

For example, a university in Fujian shared a case at the event about how it worked with Huawei Hope and its ISV partners to help the cafeteria escape from the "management quagmire".

Simply put, it provides more options from three perspectives, such as:

"Adapt to local conditions" in hardware: The Huawei Haowang D10 series fixed-focus products finally selected can be installed indoors and outdoors at any angle from 30° to 60°. Its strong universality can greatly reduce the probability of resource waste during renovation.

The algorithm layer is tailored to the individual: Based on the deep learning detection algorithm, the head and shoulder algorithm of the D10 device has an accuracy rate of over 95%, supports target detection of at least 32*32 pixels, and combines the area with the line drawing to reduce the error of wandering personnel. This can effectively support canteen management and decision-making through data collection and algorithm analysis. Students can also learn about the number of people currently dining in all canteens and their dining status through the official account, and arrange meals reasonably.

The solution is "adapted to changes": Considering the cost control needs of the university cafeteria's smart construction, Huawei and its partners in Fujian have tailored the best solution for the school, using only the D10 series fixed-focus products to achieve the solution goals of passenger flow statistics and video surveillance. Due to the maturity and high cost-effectiveness of the technical products, it has been possible to replicate on a large scale, and has recently begun to be promoted in colleges and universities across Fujian Province.

It can be seen that it is precisely because of the "capacity expansion" of the entire chain including technology, products, solutions, and cost-effectiveness that industry users can more easily plan AI projects that suit their own business needs, and the expected intelligent value can be continuously unlocked.

The second thread: AI value based on engineering foundation

Knowing how to dismantle the "tangled ball of threads" is equivalent to finding a shortcut to solving the problem of AI implementation. No matter how trivial and chaotic it is, it can quickly allow "thread leaders" from all walks of life to unravel the problem and follow the threads to break the deadlock.

For a long time, the renovation of old communities is difficult and the education costs are high, which has caused many seemingly beautiful smart solutions to run into obstacles when implemented.

For example, more and more communities have installed smart cameras during the epidemic prevention period, but after residents put on masks, ordinary portrait prevention and control becomes difficult to play a role. The professional portrait recognition solutions used in airports, high-speed rail stations and other places have increased average usage costs due to the small number of concurrent routes deployed in the community.

How can we improve the happiness and sense of security of the community without increasing residents’ troubles and cost pressure? At this time, a team with engineering capabilities is needed to enable AI with controllable costs and perceptible value to sink quickly, stably and reliably into these relatively barren digital soils.

To keep costs under control, hardware, computing power, and operation and maintenance all need to be in place. Huawei has found the root technology to solve the problem from the complex thread: an inclusive AI product with 1T computing power.

In the case of smart transformation of old residential areas, Huawei's computing power sharing pool has become a magic weapon for reducing costs and increasing efficiency. Simply put, it is the linkage of multiple cameras of D10-SIU+D10+NVR, "individual deployment, global intelligence". One smart camera as a computing power source can drive multiple ordinary cameras to have intelligent recognition capabilities, up to 64T computing power. For the same function, traditional distribution solutions may require the deployment of more than 8 devices with AI functions to achieve it.

The value is tangible, which means that the smart transformation plan can "strike hard" at the pain points in community management and effectively scratch the "itch" of residents. For example, the problem of difficult face recognition under masks mentioned above ultimately depends on the performance of the algorithm. Huawei Haowang camera uses the "super portrait algorithm" developed by Huawei 2012 Laboratory to achieve a recognition accuracy rate of more than 90% even when wearing a mask, greatly improving the efficiency of residents' entry and exit.

As for the uncivilized behaviors that are common and difficult to stop in many old communities, such as littering, illegal parking, and electric bicycles entering homes, traditional community video surveillance can only review the video after the incident, and investigate and deal with each case as it is discovered. Huawei Haowang's solution allows cameras without smart functions to monitor in real time and stop it in time.

The availability of technology, the accuracy of algorithms, and the cost-effectiveness of solutions have jointly created the engineering capabilities of AI and determined the speed of intelligent popularization. This is why Huawei's D10-SIU product has begun to be called a powerful tool for the renovation of old communities by more and more security industry personnel.