Video software vendors have a lot of solutions for the retail industry, such as applications to count visitors, determine the queue’s length and control transactions at the cash register, etc. However, there are almost no solutions for production and manufacturing. The reason for this is that video surveillance software developers make applications for production much less often than for retail. Applications for manufacturing are much more expensive. Why is that? Let's find out.
Originally, video surveillance systems were invented for security. They were designed to solve traditional problems: archive storage, operator display screens, motion detection, and archive search. Regardless of where the video software application is installed, the approach to the solution of these problems is the same. Technologies, once developed, can be used over and over again for different tasks.
Over time, video surveillance systems became more sophisticated, and now they are able to solve more complex tasks, such as detecting faces, reading license plates, and conducting searches based on various parameters. And once again, it does not matter which industry a video system is being used for, the process is the same everywhere. For the task of numbers recognition, for example, there is no difference whether the software application is installed at a shopping center gate or in a factory.
Beyond security, the next step is the solution of more narrow, specialized tasks. At first, such systems were developed for retail. They should be able to analyze the movement of customers, measure a line’s length, count visitors, and identify the most active zones of trading floors. All these functions work quite effectively.
Now it is logical to take the next step: if we managed to make software applications to analyze videos and to solve certain tasks in some areas, why not apply it everywhere? Not only in retail but also, for example, for production and manufacturing. In this case, software applications will be able to replace workers engaged in low-skilled jobs.
It is easier said than done. If basic tasks of ensuring security are rather similar anywhere, video analysis tasks for retail are also similar. It doesn’t matter what is being sold or what is stored on the shelves. Manufacturing requires a variety of tasks that need to be solved. All manufacturing facilities are different and have different production processes.
At Macroscop, we receive a lot of requests from different manufacturing companies for video analytics. For example, our customers want their video systems to be able to:
- signal the breakage of an excavator’s tooth;
- determine fractions of gravel in truck beds at a quarry;
- count the number of plastic bottles on a pallet;
- recognize precious stones on a conveyor belt.
All these tasks are very specific and so different from one another. That’s why customized development is required for each of them. Standard solutions, once developed, cannot be used for these tasks. But customized development is expensive.
Naturally, a product produced in larger volumes is cheaper than a product produced in smaller volumes. In order to develop technology and begin production of a product, it is necessary to invest a certain amount of money, which will be recouped through sales of the product. The greater the number of units of the developed product sold means smaller the amount of production costs per unit. When a solution is designed for a unique customer it is produced as a single copy. The probability of another customer buying the product is rather small. The whole cost of the product’s development is apportioned to its price. A simple rule applies: lesser volume means greater price.
Ways to save money
At Macroscop we know examples of non-standard task solutions which were achieved with minimal costs and utilized for manufacturing. These companies used creativity and applied existing standard tools of video analysis.
One of the roofing materials manufacturers uses a video system for defects (holes) detection in produced materials by use of a motion detector.
The materials produced are pulled through a dark box on a conveyor belt. At the bottom of the box, there is a beam of light. It is directed vertically up (through the tape). A camera is installed at the opposite end – on top - and is directed vertically down. If there are no defects (holes) in the material, the picture on the camera is completely black. But if there are some defects in the material, the camera captures beams of light. Macroscop is using this approach to configuring a special external application. The application uses a motion detector, which captures light beams and notifies the operator of the detected defect.
The software component cost for this solution is the cost of license just for one camera.
Alternative point of view
Our point of view is not shared by everybody. Some of our colleagues believe that all existing tasks, no matter how different they are, can be classified and put into certain categories for the purpose of universal solutions development.
A key success factor with such an approach is that tasks grouped together should be very similar to each other. If that is the case, it makes it possible to use a similar approach to effectively resolve their solutions. But if a group contains five tasks which seem similar, but, in fact, have no universal algorithm, such classification becomes useless. For example, what can you group with a task detecting whether a tooth on an excavator is broken?
On the other hand, it is obvious that this approach will be used in the future. It is possible to classify tasks into groups based on, for example, the recognition of a certain subject or a certain event. Hopefully, in the near future, it will be possible to use certain video analysis algorithms in order to recognize absolutely anything. With some limitations, such algorithms already exist.
Not so long ago we met with a company named “Blippar”, which, as they say, created the first visual browser. They made an app for mobile devices which can detect a specified item captured by a mobile phone camera and give output on the content of this item. It is a workable application already which shows pretty good results.
Summarizing everything said above, we may conclude that in video analysis there are already some very promising technologies. These technologies are universal and can be applied to different tasks without major alterations.
It is obvious that the task of recognition of certain subjects or events will become fundamental in the future. This core functionality can be applied everywhere, even including customized tasks for production facilities.
But we’re not there yet. At the moment, while technology hasn’t reached such heights yet, manufacturing companies with narrow, specific objectives have three options at their disposal:
- Use costly applications developed on an individual basis.
- Be very creative and utilize standard tools and solutions for addressing non-standard tasks.
- Substitute video analysis with human resources.