Introduction

The objective in this step was to perform hand detection (an essential objective to detect the picking) directly from a video camera, with which I made a prototype capable of reading from any video camera on the computer or capable of charging a Youtube or disk video and perform the analysis on the video.

Hand detection

This should be the way to detect the picking (or at least one of the components to make this kind of defection). Remember that the final goal that I have set is to build the software for a SmartShop and therefore “hands” are one of the important problems to solve. Once I managed to get a trained model to detect hands I have achieved very good results. The following example comes out of the processing of a Youtube video. The video does not have good input quality, but anyway this is the result. From this source video:  

We will get this result:

Results

The results are pretty good, so the next step will be to identify the people entering the store and to link the person with the identity.

Technical details

For those of you who may be interested in the technical details, here you have them.

Hands detection

In order to execute this task, the first was to look for a database to start from (one thing is to fight with the development and concepts and another is to shoot and photograph thousands of hands to be able to train the model). The database that has given me better starting results is EgoHands (A database of hands with a self-centered interaction), this database can be found on the Indiana University WEB. You can find the source code of a sample that allows you to start to work in things like this en my Github repository https://github.com/SignalRT/DemoTFObjectDetection