It is possible to combine machine learning with DJI drones to provide real time object recognition. This early example shows a phantom 4 pro drone flying over a suv and successfully recognizing the suv. Also the landing pad is occasionally recognized as a frisbee.
What is exciting about combining machine learning and DJI drones is the ability to easily recognize custom objects for specific use cases. For instance my next project is to recognize blackberry bushes in mint fields to allow local farmers to locate and remove weeds in their fields. It is now easy to create custom object detection systems and then include that in an android app and to use this with a low cost drone such as the phantom 4 or mavic.
Currently it takes 1.5 second to analyze a frame and detect the objects. However future development will include methods such as tensorflow lite models which google specifically designed for mobile devices and should make the process much quicker.
Another method to increase the analysis speed is to use a faster android device. In this instance a DJI Crystal Sky device was used. In the future as faster android tablets are released specifically to quickly process machine learning routines then this will speed up the real time analyse.
Share on Facebook
Share on Twitter
This blog goes through the steps to develop an android ground station for DJI drones called DjGs. Developing applications for drones is still in its b...