Computer Vision Systems with AI Agent are mainly used for:

Read more
Read more
Read more
Read more

CAFA Tech develops and adapts AI Agents for Computer Vision tasks to help companies with common problems with the ability to add customized functionality as needed.

Main benefits of Computer Vision System with AI Agent:

  1. to enhance work and industrial safety
  2. to reduce human workload regarding video analysing, inspection and repetitive operations
  3. to increase productivity as AI helps with the analysis of efficiency of work processes
  4. to increase quality thanks to automated analytics

Software Development for Computer Vision

GPUs are the state of the art in training and working with deep neural networks. In 2020 completely self-learning systems are yet to be seen. However, prepared datasets of images can be input for the CV learning algorithms to train artificial neural networks for various object detection algorithms. This method- human-guided Computer Vision Systems- ensures that learning process is strictly under control and results are predictable.

Computer Vision System can be build using different approaches:

  1. Cloud based platforms Google, Amazon, IBM Watson etc. These are universal nodes for various tasks of AI but the software for them needs to be developed, configured and tested before use.
  2. Specific software, for example: Logistics, Construction site monitor etc. These softwares still needs to be developed, configured and tested for each company.
  3. Company’s own Computer Vision System with AI Agent. In this case, company has own AI Agent which is adapted to help a company specific tasks. Development and mentoring service can be outsourced. The hardware may be in-house or use computing resources provided over the Internet.

Some modern IP cameras have built-in AI for face or vehicle detection but these functionalities are restricted by limited computational performance. One cannot expect the same computing power inside the IP camera on a small chip or a server with a GPU unit. Server based AI Agent has much more flexibility regarding functionality, computing resource allocation and upgrades. Also, the options for IP camera brands and models are not as restrictive.

UAV (Unmanned Aerial Vehicle i.e. drones) and UGV (Unmanned Ground Vehicle) Computer Vision solutions

Computer Vision solutions consist of hardware and software. From the hardware side, it is important to consider the requirements to make particular analysis possible (in real time or not).

Real time and close to real time analysis requires specific hardware with specific parallel computing capabilities. Modern solutions often use GPU hardware for that.

It depends largely on the analysis type and optimization but for reference one modern average GPU card (11 TFLOPS) can run modern object detection algorithm on Full HD video resolution 30FPS.

UGV (Unmanned Ground Vehicle) usually has 4 on-board cameras to cover 360 degrees. With the same frame-rate for all the cameras, the on-board processing capability would have to be roughly 44 TFLOPS.

Computer Vision Systems for PPE work safety and for industrial safety

AI based video analyzer for industrial and robotics safety

CAFA Tech (Technology Provider) will develop and set up the Industrial and Occupational Safety AI based computer vision solution (CAFALYZER).
CAFALYZER provides computer vision for automated situation assessment:
1. Inspection of the workers wearing PPE.
2. Checking machine tools when protective sensors are covered.
3. Checking the safety of handling marble slides
Adjustment of computer vision alert thresholds and performance of specific functions is possible when a work safety coordinator changes settings of CAFALYZER.
The Novelty solution is based on the fact that CAFA Tech provides video analytics coaching by a corporate mentor. Mentors engaged in industries help to achieve industry specific and partially universal AI development.
It is good for the company to have its own AI trainer and personnel instructor, as fully self-learning systems are not yet world-class to learn from the surrounding world without human guidance.
The use of trainers specializing in safety analysis helps ensure better analytics with CAFALYZER.
Computer Vision artificial neural network weights are calculated by dedicated hardware and software on a workstation but it requires a large dataset of specifically prepared images for that. Raw images can be aquired directly from the stationary cameras with some coordination from a work safety expert and AI expert taking reference images from actual workers and machinery.

Dataset image preparation (for machine learning) can be done by an expert from a remote location (remote work) and can be even performed by a person with physical disabilities.
Companies want to outsource this service as a complete solution. CAFA Tech offers a complete solution: planning and installing cameras (with the help of security companies) and training and upgrading an AI agent for company.
Step by step, a customer company can move on to the following and more sophisticated services, where the AI agent can also evaluate ways to increase production efficiency.
Developing AI algorithms and 3D planning know-how and working with companies is a key feature of CAFALYZER.

Computer vision (CV) based image analysis must be set up and running continuously to provide close to real time detection of PPE related issues and notification of employees of the detected problems. Data collection requires mapping of hazardous locations and installation of video cameras. According to the number of video cameras, the computing power must be planned in order to achieve near real-time analysis results. Audible and visual alert options must be added to the monitoring workstation if the CV AI detects a hazardous situation.

The Novelty solution is based on the fact that CAFA Tech provides video analytics coaching by a corporate mentor. Mentors engaged in industries help to achieve industry specific and partially universal AI development.

In 2019, metal processing accidents in Estonia had the highest number of occupational accidents (483 cases). This was followed by industries: woodworking (338 cases) and then the food industry (203 cases).

CVS to evaluate the efficiency of processes:

  1. Analyze the trajectories, speeds and delays of mobile objects and personnel to find optimization options by changing the layout of the work area or changing the timing.
  2. Analyze congestions due to less than perfect ad-hoc storage areas.
  3. Analyze the safety regarding storage use
  4. Find time-critical logistics problems that influence the overall production efficiency the most
  5. Analyze trends (cluttering, contamination of surfaces, build up of storage area)
  6. Analyze overall employee dedication and behaviour

CVS to control quality of products:

Visual inspection by human(s) is not fast nor reliable enough by modern standards. As long as products can have known types of defects, it is common to use dedicated machine vision apparatus for detecting them automatically. When this quality control apparatus detects a defective product, it can quickly send command to an actuator, which removes the product from the production line to a container that can later be inspected by humans with no rush. As soon as the rate of defects goes above permitted threshold, the production line can be stopped automatically for further investigation by trained personnel.

High level workflow Computer Vision System (CVS)

Implementation of company CV System with AI Agent

See how CAFA Tech can help you

Read more