Sumo real time object classification with faster rcnn


If anyone is interested I’ve done a project which does object classification of a sumo video feed real time. I’ve got a youtube video of it in action at:

The Sumo is being controlled from my PC which in turn is using Faster RCNN to process images real time putting a box around glasses, mugs, shoes and apples. Currently I have trained it for these four objects only. It was trained using faster rcnn deep learning from over 600 images. If you would like to run the project yourself the project is on github at:

It is a fork of py-faster-rcnn. To run it you really need a PC with a NVIDIA GPU that is 960 or better if you want real time processing. You will also need to install CUDA. It is also much easier to compile on Ubuntu.


If anyone at Parrot is interested in help developing object classification with Parrot drones I’d be interested as it is an area I’m keen to get into. I’ve spent a lot of time with object classification with most Caffe implementations such as FCN segmentation, faster RCNN and normal caffe single image classification. I’ve trained many original models for these caffe implementations with original training images. Also currently an area I’m exploring is using an NVidia Jetson TX2 with drones to allow on board image classification. At present my object classification with drones is done by processing images sent from a drone via WIFI to a second computer.