Machine Vision is the ability of an AI to recognize objects visually. The AI is trained for specific scenarios and becomes proficient with usage, as data experience increases. Each scenario becomes a ‘skill’ learnt by the AI.

The core applications for Machine Vision include counting and visual recognition of objects. The analysis is based on the specific learning model the AI is trained for – for example detecting road traffic.

A composite Machine Vision skill, intrusion detection relies upon recognizing a range of targets, including people, animals and vehicles. This skill is often used in exterior surveillance and security scenarios.

Counting can be utilized in two distinct ways. The AI can be asked to do an ad-hoc or ‘freestyle’ count based on the images or videos. Or, it can be a structured count, where double-counting is removed by supplying additional metadata such as GPS coordinates. This method is used for orchard yield estimates, livestock tracking and so on.

Counting can be also be utilized in real-time. This always involves analyzing a live video stream with live alerts and annotations. Advanced tracking algorithms enable a structured approach and this technique can be used in counting crowds, estimating wildlife populations and more.

Historical data when categorized properly, can be used to build a ‘stack’ that can be analyzed utilizing other types of Machine Learning frameworks. This comparative analysis allows the AI to extract hidden patterns from the data stack, which leads to better decision making.

Based on Pix4D, this service delivers mapping and surveying outputs, including orthomosaics and elevation maps, from geo-tagged imagery. The input data must conform to the minimum standards required for surveying.