Object Detection

Using tensorflow framework to build a deep learning model that will detect hand signs from web cam and surround it with a bounging box.

Data Collection

The following train and test sets were created using OpenCV and labelImg.
Train set:

Test set:

Model Used

From Tensorflow Zoo, the pretrained ssd_mobilenet_v2_fpnlite_320x320_coco17 model is used. Installed using Tensorflow object detection API in Tenserflow Model Garden.

Training and Testing

Model was trained using the default parameter and 2000 steps.

Loss graph in Tensorboard

Result on the test set.

Expected Outout in Tensorboard

Packages

Future Scope

To make a web app that detects hand signs making webpage browsing interactive.