Learn about reference implementations for retail spaces and take a closer look at the Store Traffic Monitor.
Hello. My name is Martin Kronberg. And this is the IoT Developer Show, season three. In this episode, we learn about the reference implementations Intel created in the retail space and take a closer look at the Store Traffic Monitor, a retail solution for—you guessed it—monitoring traffic in a store.
Retail stores are evolving. With such competitive markets, retailers want to capture people's attention and make shopping experiences more efficient and more fun. As a result, it's more common to see smart displays, automated kiosks, and reactive advertisements in retail spaces. Many of these new retail solutions are based around computer vision. Intel has created a series of reference implementations that can be used to teach the concepts of leveraging computer vision in the retail space, as well as provide you with an open source solution that can be used in creating real world customer applications.
Right now, there are seven implementations covering things like face access control, gaze monitoring, and store traffic monitoring with more to come. Let's take a closer look at the Store Traffic Monitor right now. Welcome to our virtual store. As you can see, this solution is able to track people in the store as well as specific goods, in this case, bottles. The Store Traffic Monitor is also able to keep a timeline of when people arrive, how long they stay, and when they leave. This is all pretty useful information to have if you are a store manager and want to do analytics on when customers are most likely to come in and what products they consider buying.
This solution is running on an IEI TANK* AIoT Developer Kit with an Intel® Core™ i7 processor and uses the Intel® Distribution of OpenVINO™ toolkit. There are two versions of this solution— C++ and Python*—so you can choose the language that you're more familiar with. This solution is able to either use prerecorded video or a live camera stream. To do the person detection, the Store Traffic Monitor uses a single shot detector or SSD. You can use any model that has the input and output format of an SSD.
Some models are trained to detect different objects, some run faster or slower. So you'll have to find one that fits your specific needs. You can use the prebuilt models straight out of the box or any SSD Caffe*, TensorFlow*, or MXNet model by utilizing the model downloader and optimizer included in the Intel Distribution of OpenVINO toolkit. Keep in mind that you'll have to make sure to have a corresponding labels file for each model. This is a file that tells you, say, object 15 is a human and object 30 is a bottle.
For this demo, we're using the MobileNet SSD model and the labels file is included with the project in the GitHub* repo. For more details about how this reference implementation works, be sure to read through the GitHub repo and source code. And that's a Store Traffic Monitor reference implementation. I hope you have a good understanding of how this solution works and some of the ways you can customize it for your specific project. In the next episode, we learn more about the Intel industrial reference implementations [sic] and take a closer look at an object flow detection solution. See you next time.