The technology to track people requires three protocols: detection of objects, recognition of the object as a person, and tracking the object across distance and time.
Detection of Object
What’s Detection? It means we detected an object in a specific location. The location can vary in size from an object in a zone to a static pixel. Either way, the concept of detection is binary.
Either we detect the object or we don’t.
In technologies such as video, thermal, and Time of Flight, we aim to detect all the objects within an area. If we miss an object, the system under-counts. The scientific name for under-counting is Recall.
The other part of the equation is the accuracy of the detection. With Precision, we measure if all the detected objects are people. The difference is over-counting.
For example, there are 20 people within the sensor’s Field of View. If the solution counted only 18, the under-count is 2. And if within those 18 object, 1 object was not an actual person but a shadow, then we over-counted by 1.
The relationship between Recall and Precision is the accuracy. The Accuracy Rate is the balance between under-count and over-count.
The scientific name for Accuracy is the Threshold Rate.
A heat map is an example of detection. The red dots are the detected objects. If we missed some people (no red dots), we under-counted. If some of the red dots are not people but carts than we over-counted. The heat map is a visual presentation of detection.
Recognition of Object
The concept recognition means we identify the object. There are different levels of identification and the technology is getting better.
In people counting, the criteria are the shape of the object. Video counting, for example, depends on the recognition of a person by their head. The recognized object is the top of the head. On the other side of the scale, we have detailed attributes of the object. With Facial Recognition we can recognize a specific individual.
With images, the market for recognition of is evolving. One factor is the detection required for people tracking. We deploy image recognition to identify of the products are on the shelves. Image Analytics is also used in driverless cars and Facebook Ads.
In retail, there are three cases we use object recognition.
One is for surveillance and loss prevention. Another is for compliance of stocking. The most common, and pernicious, usage of facial recognition is in profiling and marketing.
Once we detected the object, and recognized it, the next step is tracking.
Tracking of Object
Tracking relates to video images, radio signals, and other objects. With tracking, we should know where the object is, but also where is it moving towards, and how long it will take to get to the next point. This combination of location and time is the idea behind motion of objects.
For us, tracking is the concept of people in motion. And the accuracy of people tracking relates to motion. People Tracking, therefore, has its own kind of accuracy errors. In addition to errors in detection, we have errors in Switching and Precision.
ID Switching means we continue to track the same object (person) along the path. Precision is the predictive component of forecasting where the next step is going to be. In essence, tracking is about prediction. The more, correctly, we predict the next move of the object, the more accurate is the technology. In other words, tracking accuracy depends of the precision of the prediction.
Let’s look at two scenarios of WiFi Tracking.
In a supermarket, we capture a signal from a person in the bread aisle. 30 seconds later, the same signal is in meat freezers location. Where do you put those 30 seconds, in the bread aisle or the meat freezer seconds? These events skew the analysis of customer engagement.
A second example is from a Department Store. When customers enter the store, they first stand in Cosmetics. The next time, we capture the signal in the Baby Section. But the path can carry the same person either though the Women’s Dept or the Men’s Dept. We have no idea which path the customer had chosen. We only know the person started in Cosmetics and ended in Baby Department.
Bringing It All Together
People Tracking depend on the ability of the technology to detect, recognize, and track objects. The accuracy of tracking depends on predicting the motion of objects.