Key considerations when choosing cameras for edge AI scenarios

Key considerations when choosing cameras for edge AI scenarios

Here at Neal, we approach computer vision scenarios with a production mindset. From the edge to cloud infrastructure to the physical hardware implementation, we want to be sure the solutions are designed with production scale and performance in mind. As such, we often assess our customers’ environment with each specific scenario to determine the best camera selections to meet their needs. As part of this process, we must consider several areas to inform the camera selection process.  

Consider the scenario requirements 

Each edge AI scenario will have its own set of requirements for camera placement, IP ratings, network connection, and more.  

Key considerations for edge AI scenarios

1. Indoor vs. outdoor

For example, a key question to start with will be: Where do you need the analysis to take place?   

If this scenario is taking place outdoors, you will need to consider ingress protection (IP) ratings and tamper ratings for your hardware. IP ratings measure resistance to dust and water exposure. IK ratings measure resistance to physical force (frequently used for vandal resistance.)   

On the other hand, if your analysis takes place indoors, you have more flexibility. These cameras will be less exposed to water, weather, and physical force.  

2. Expected conditions

In addition to being indoors or outdoors, you’ll also want to consider the environment where this analysis will take place. What type of conditions to expect to encounter?   

This is particularly important when looking at lighting. For example, does the solution need to perform in the low light of nighttime conditions? If so, you will want to consider infrared capabilities.   

3. Camera placement 

Camera placement will be critical in your edge AI solution, so it’s important to map this out ahead of time, as other considerations, such as lighting, weather, or network connectivity, will have an impact.   

When considering optimal camera placement, the Neal team looks at three key areas 

      • Area to be monitored 
      • Camera distance from the area 
      • Mounting options 

Area to be monitored  

Let’s start with considerations around the area that will need to be monitored in your solution. If a large area needs to be monitored, you may want to consider cameras with motorized pan, tilt, and zoom (PTZ) capabilities. In these cases, a PTZ camera may be able to cover more territory and potentially cut down the total number of cameras required.   

If the cameras only need to monitor a limited area, then a fixed-position camera will be adequate.   

Generally speaking, PTZ cameras cost more than fixed-position cameras, so you will need to consider the tradeoffs between fewer cameras and increased flexibility. Additionally, using a single camera for a larger region means it cannot constantly monitor all regions simultaneously. If you need continuous feeds for all regions of interest, then fixed-position cameras are likely a better fit for your scenario.   

Camera distance   

Another factor in camera selection will be distance. How far can you feasibly place the cameras from the area that will be monitored? If the distance becomes an issue, then optical zoom will be an important feature to have. Without an optical zoom lens, you should purchase precise cameras with adequate focal length, given the exact use case.  

Mounting options  

Finally, where will these cameras need to be mounted to effectively monitor the area?   

Does the camera need to be placed on the ceiling to get the desired angle? Could it be placed on a wall? Are there any other camera mounting constraints, such as drop ceilings or existing power or network reach limitations?  

Understanding these constraints will allow us to select a camera and assess the available mounting harnesses. These can range from pendant mounts to wall mounts, extensions, and more.   

4. Network connectivity 

Moving back to the requirements of your edge AI solution’s operating environment, it will be important to note any possible networking constraints. If this environment is heavily constrained, you may want to consider an integrated camera. Integrated cameras are low bandwidth cameras, making them a good choice for settings like low power or offline batch processing, where network connectivity is often an issue.  

Or does the placement support an ethernet run? If that’s the case, you’ll likely want to run the power over ethernet to power and connect the camera with a single cable. If not, you’ll want to assess the feasibility of Wi-Fi-connected cameras. 

5. Frequency of analysis 

How often will you need to run the video feed through your AI algorithm for analysis? Some scenarios will require inputs several times per second, while others can have the required business impact with much less frequent inferencing. This will drive the required framerate or FPS that the camera will need to support.  

Consider the industry norms 

Different industries have different uses for their cameras. 

6. Industry uses for camera

Two primary form factors are present: IP cameras and machine vision cameras. 

IP cameras 

IP cameras, for example, are great for retail and security scenarios. These cameras most commonly use the real-time streaming protocol (RTSP) for connectivity.  

Something to look for when selecting IP cameras is Open Network Video Interface Forum (ONVIF) support. ONVIF provides a variety of standards that cameras must conform to in order to be certified. Typically, the Neal team works with cameras that have ONVIF profile G, S, or T compliance. Some example vendors include Axis, Bosch, Hikvision, and Hanwha.  

Machine vision cameras 

While IP cameras are commonly found in retail and security, machine vision cameras are more common in the manufacturing industry and are often placed on the production line.  

One thing to note is that the lens and camera are typically purchased separately. The lens must be selected based on the required field of view, depending on where the camera will be mounted. Another consideration for lens selection is the camera sensor size, which will determine the field of view.   

Machine vision cameras use the GigE vision streaming protocol (GVSP) for connectivity. Some example vendors include Basler, Hikvision, and Allied Vision.  

USB cameras 

A third, less common form of camera is a USB camera. In some cases, a USB camera or webcam will be a suitable input device for running the solution. This is especially true if it’s in a fixed position at a desk.   

Consider the existing business environment 

When looking at these camera and hardware considerations, it’s worth noting that not everything has to be done from scratch. Sometimes customers have cameras that can be used for their solution, or they wish to leverage existing relationships and vendors to stay within a certain ecosystem.  

Existing cameras 

Do you already have cameras in place that could be leveraged for your edge AI scenario? 

In some cases, customers can leverage their existing cameras if they adequately meet the considerations listed above. However, you may find that new cameras are still required because the existing ones don’t match the scenario criteria. For example, the existing cameras may not support the required viewing angle for the scenario.  

Additionally, cameras used for security purposes are often off-limits, even if they are adequate for the edge AI scenario. This is typical because security teams have dedicated them to other surveillance purposes. We recommend checking with your infrastructure security team to determine whether new cameras will be required.  

Relationships with OEMs and vendors 

Are there existing relationships with original equipment manufacturers (OEMs) or preferred camera vendors within the business? 

Sometimes we’ll find our customers have existing relationships with camera OEMs and that they would like to remain within the existing ecosystem. In these cases, we can use the inputs gathered from all the considerations above to inform our camera and hardware selection from the OEMs catalog.  

In other cases, we are happy to recommend a few camera alternatives that would be suitable for the scenario. Customers can procure these internally, or we can procure these in bulk for you as required.

Next steps 

All these considerations around edge cameras and hardware can seem daunting. That’s why Neal Analytics helps customers select the best edge cameras and hardware for their solution, business setting, and use cases.  

We have deep expertise in IoT & Edge solutions and can leverage our years of experience to help you streamline the process. 

Contact us to learn how our team can help you move forward with your IoT & Edge solution! 

Learn more