We make a bold claim: Entrance is a simple solution for a seemingly unsolvable problem.
Our software only needs a survey of the entire entrance area to do what previously seemed impossible: Make sure a retail store’s open entrance area is not being exploited by shoplifters as an unmonitored exit, without putting a gate in front of it or scaring away honest shoppers.
With every claim to simplicity comes a gut reaction to disbelieve it.
Rest assured: Using Entrance is simple. But, a lot of work on our end goes into making that happen. Luckily we’re able to do that work within seven days for each installation.
There are three main steps required for an Entrance system to reliably recognize pushout attempts. Today we are taking a look at them.
We won’t be giving away all the magic. Still, our claim will seem less bold after reading this.
Here’s a demonstration:
The moment the screen flashes, Entrance has identified a potential pushout attempt.
Step One: Calibrating the cameras and defining zones
We make existing CCTV cameras smart. First, the cameras need to be calibrated to monitor the entire open entrance area. This does not require excessive training. A careful look from either one of our partners or our in-house installation team is needed though.
Entrance makes gates unnecessary. To do so, our customers define zones within the entrance area. Within the ‘good zone’, carts and baskets are allowed to move freely. The ‘bad zone’ – defined by a simple line – is where the magic happens.
At this point already, our customer’s work is done.
Step Two: Learning to detect and track Carts
The requirement for tracking carts and baskets is identifying them as such. This is a trivial task for humans. Computer systems, however, need a lot of examples to learn how to do it. Carts and baskets differ slightly from store to store. Thus, we need to do a lot of work to get the best results possible.
Our models are trained using curated data. We pick examples that teach our system what carts look like.
Our team of annotators spends hours looking at security video feeds, identifying and labelling carts and shopping baskets. The results are used to train an intelligent model optimized to detect carts and baskets in the respective store.
Step Three: Classification of carts
Imagine an alarm sounding, whenever a customer pushing an empty cart or carrying an empty basket turns around because he forgot his wallet. Would it deter thieves or customers?
Entrance is more sophisticated than that. We train a model to detect each cart’s and basket’s fullness level, which again requires hours of annotated video data.
Once the model is ready, Entrance only has one task: Track every cart and basket as it moves through the zones. When a loaded cart enters the ‘bad zone’, the system is triggered and reacts appropriately.
Let’s look at the demonstration again. This time we’ve added some visual cues, to help you make out what’s going on:
With a lot of successful installations under our belt, we have models ready that can handle most store set-ups. Still, in every single case, training is required to produce the best results possible.
Entrance is affordable, elegant and easy to set up and use. It is a simple solution. We take care of the hard part.