chinese-chess's Introduction

Chinese Chess Recognition

To solve this problem, the first step is to identify the beginning point and the end point of the pieces movement. The second step is to crop a square image on the begin/end point and pass it to convolutional neural network (CNN) to do the chess pieces classification.

Test Video

Usage

CNN_Classification_Model contains the codes I used to train the classification model, it is ok to use the .h5 model named new_model_v2.h5 in the h5_file.
Dataset contains the dataset I made by taking pictures by Phone and using HoughCircle to roughly extract some chess pieces from the picture.
The .h5 file model works for those images in the Dataset with nearly 100% accuracy, if it is used to detect other kinds of test image with different light intensity or different size, the accuracy may get lower.
Temporary_Model and Test_Image are two necessary directory used in the codes.
AdjustCameraLocation.py is used to adjust the camera to maximize the area of the chess board in the picture.
real_time_test.py is the main function of this project.

Notes

The training data in Dataset/train lost some images because of some unknown reasons, if you need to re-train you model, you can generate more data by yourself or just move some data from valid to train :).
The location of the phone need to be right over the chess board, it is very hard to fix it (I used the mobile phone holder like this). So I provide a video named test.avi in the Sources directory. If you want to do the real time test, you need to change the code in line 240, real_time_test.py.