IAS edge AI detects and tracks defined objects. When the object enters the relevant zone, the IAS alerts, such as the display and external alerts, are triggered.

IAS edge AI runs entirely on the Processor on the vehicle within 150 milliseconds and does not require any internet connection to work.

IAS has 3 user-selectable objects:

IAS can be configured to detect a single object type or all objects at once. There are also several non user-selectable objects so that the AI does not mistake them for one of the user-selectable objects.

IAS edge AI uses the YOLO (You Only Look Once) framework, based on Deep Convolutional Neural Networks (D-CNNs).

The IAS edge AI has been specifically trained on heavy industry sites over many years, leading to unrivalled performance.

A false positive is detecting an object as another object. For example, the below image shows an AI detecting drain ends as a person.

A false negative is not detecting an object that is present. For example, the below image shows an AI not detecting the third person.

A false positive is generally considered ‘safer’ because it detects 100% of the objects present, at the risk of being annoying to operators when there is no object there.

However, some objects, such as tires, train wheels, and hi-vis clothing, defy easy categorisation. For example, the below image shows a hi-vis jacket being detected as a person because within 150 milliseconds even an operator cannot determine if it is a person or a jacket.

The IAS edge AI will generally take the more cautious position and detect them as people/vehicles, increasing the number of false positives.

AI performance depends on many factors, including the vehicles, environment, and other objects. It is impossible to provide an accurate estimate of AI performance before being deployed on site.

A good rule of thumb is the AI will be greater than 90% correct out of the box and improve throughout the initial deployment with the automated over-the-air updates.

Edge AI issues can be reported by users in IAS Reporting. Users can mark the detection as an AI error and provide some additional information. This is automatically sent to the AI developers for review and to update the AI.

All IAS users therefore benefit from this review and AI update process.

Note that no picture or video data is ever sent to other IAS users.