What is Edge AI?
Edge AI is all about running AI algorithms locally on a physical device collecting data on-site. Avinton’s Edge AI Camera is a highly customizable solution utilizing the latest technologies in AI, machine vision and edge computing.
Edge Computing x AI x Machine Vision
With Edge AI, inferences can be made on the spot, paving the way for real-time business intelligence. AI is on-premise, unlike other options like Cloud AI where the model is secluded in a separate environment.
The solution can be applied to a wide range of real-world use cases, ranging from the securing of worker safety at production sites, to acquiring of customer attribute information for marketing purposes.
Why Edge AI?
Edge AI enables complex tasks to be performed right on the spot – at high speed and low cost. Our Edge AI Camera consists of a nano-computer on which the trained model is deployed.
Real-Time
Time lags are an inevitable consequence of AI processing in the cloud; where safety is at stake, milliseconds matter. With Edge AI the collection and inference of data are performed on the same device, making it up to 10x faster than cloud AI solutions.
Low Cost
Whilst the one-time setup cost of Edge AI is just slightly higher, the long-term costs can be upto 8x lower than cloud-based solutions. Reason: running heavy computation in the cloud can be very costly, especially if your device is running 24/7.
Versatile Design
Attach any USB / MIPI camera to suit your need and budget. Multiple cameras can be attached to the same hardware. Control switches, relays and environment sensors can also be integrated depending on the environment. All in all, the solution is highly customizable.
Small Size
The nano-computer hosting the trained AI model starts at just the size of a credit card. There is no need for securing a dedicated workfloor space for the installation; the device can be placed inside existing equipment control boxes.
Low Power
The device runs on low power, enabling it to be installed in secluded environments. Starting from 15W, it can also be powered by solar using battery backup.
Edge AI is a game-changer
The traditional approach to integrating AI with optical devices like camera is through cloud computing. Data collected on-site is sent to deep learning models residing inside cloud-based data centers due to the need for heavy compute capacity. There is a reliance on network connectivity and stability, since inference on the data is performed in a remote location.
The utilization of edge computing technologies brings AI to the very edge of the network i.e. the optical device itself. A nano-computer within the Edge AI Camera will host a trained AI model to perform complex tasks right on the spot – much faster and at lower cost.
Cloud AI vs. Edge AI
| Cloud AI | Edge AI | |
|---|---|---|
| Camera <> Cloud Bandwidth | High | None |
| Cloud Compute Cost | High | None |
| Scalability | Low (costly) | High |
| Running Cost | High | Low |
| Camera Cost | Low | High (but not by much) |
Edge AI Camera - Features
A one-size-fits-all approach does not make sense with AI. Our solution offers a wide range of features that can be tailored for users across industries. Whether it be to secure safety in the workplace or improve marketing practices, we will prepare just the right model and make it “edge-ready”.
Object Detection
Object detection is used to detect instances of semantic objects of predefined classes, whether it be people, vehicles or pieces of equipment. For example, a device running a model trained to detect hands can prevent potential injuries and accidents when using heavy machinery at manufacturing sites.
Defect Detection
Surface anomalies such as cracks and dents can be detected with models trained specifically for the purpose. For example, a device running such a model can be placed on production lines to detect tiny defects to augment the precision of manual inspections. Such solutions can greatly improve quality management.
Posture Estimation
The posture of an individual can provide valuable information about his or her status. By detecting key feature points and connecting them together, such a model can estimate postures with high precision. Postures can indicate suspicious or potentially dangerous behavior that can be prevented upon their detection.
Fall Detection
Slips and falls are indicators of severe safety risks. By extending the capabilities of posture detection, this feature can increase safety in certain environments, such as healthcare institutions where patients falling over can be detected for immediate attention.
Access Detection
Traditional surveillance cameras simply record footage to be monitored later. By running a model trained to detect people, real-time detection of unauthorized access to restricted areas is possible, thus leading to stronger security in your environment.
Object Tracking
Models can be trained to track the direction of moving objects and people. This feature identifies visitor flows and popular routes, allowing you to improve point-of-sale marketing and traffic management.
Object Counting
Counting objects manually can be time-consuming, costly and hugely inaccurate. Keeping count of the occurrence of specific objects is the most efficient way of quantifying them in your environment, whether it be the number of visitors to a certain facility or the number of passing vehicles.
Gaze & Emotion Detection
An individual’s direction of gaze can provide valuable insight into their point of interest, and their expressions reveal their sentiment towards that object. Such attributes can be very difficult to collect with traditional manual methods. Use trained models to enhance your marketing strategies.
Edge AI Camera - Use Case
Issue
Worker safety at manufacturing sites
Accidents in the workplace are especially persistent in the manufacturing sector, and the number of incidents is on the rise. Cases where parts of workers’ bodies are caught within heavy equipment have severe consequences on the well-being of workers, and therefore the securing of safety is of highest priority for any company.
Avinton’s Edge AI Camera is now in the process of being deployed in a leading company in the field of printed products, industrial supplies and electronic devices. Specifically, it has been introduced to the manufacturing site of packaging for foods, beverages, household items and pharmaceutical products. Due to the nature of these products, such sites are riddled with rotating equipment that could lead to potential accidents.
Secluding areas of risk by fencing them off may actually increase the risk of workers getting caught in the equipment as their movements are restricted. The use of thermal sensors to detect workers near equipment was not a realistic option, due to the fact that they have protecting clothing from tip to toe.
Approach
Detecting workers with Edge AI Camera
Avinton’s Edge AI Camera utilizes the latest technologies in Edge Computing, AI and Machine Vision. It can be fully tailored so that it is meeting the specific requirements of our customers.
Following several on-site verifications, we are now developing a system centered on the Edge AI Camera which detects both the hands and body of workers entering a pre-defined “area of risk” on the factory floor.
When securing safety in manufacturing plants, milliseconds matter. Since the collection of data and its inference are performed on-site, there is hardly any time lag being generated within the process. The size of the Edge AI Camera is small, meaning there are very little restrictions when installing it. In addition, the fact that external switched and sensors can be attached to it means that it is easy to integrate the Edge AI Camera with existing equipment.
Why Avinton?
Strengths
Avinton has been handling AI-related projects in various domains, from manufacturing to pharmaceutics. Our strength lies especially in the integration of AI with machine vision, as encapsulated in our Edge AI Camera solution.
All of our development is done in-house; we never use subcontractors. Therefore, we are confident that the solutions delivered to our customers are always of the highest quality and reliability.
We have a team of engineers rich in both age and ethnic diversity, with expertise in various domains of IT. This enables us to tailor our solutions to individual customer needs, and integrate them into production-ready systems.
Consulting Services
Avinton is involved not only in the development of IT solutions, but also the provision of technical consulting to various clients, from small companies (50~100 employees) to multinational enterprises and local governments. The team is managed by our lead consultant with over 15 years of experience in more than 10 different countries. Some of our AI / Big Data related consulting services include:
– Custom Data Connector Development
– Custom Data Pipeline Development
– Data Visualization Application Development
– AI Modelling & Training
– PoC for Data Verification
– Technical Training (Avinton Academy)
Contact Us
We will be happy to provide you with more information.