This model has been built to classify images as No ball or Legal ball
Design an algorithm to detect whether a bowler delivery is a Legal or No-Ball delivery using the images of bowlers in action. Our goal is to measure the probability of an image being a no-ball or not and to make the automated umpiring system and to eliminate the shortcoming of human perception.
We have deployed a Convolution Neural Network (CNN) based classification method with VGG19 to automatically detect and differentiate foot overstepping no balls from fair balls. We have used Transfer learning algorithms which uses the knowledge gained from solving one problem and applying it to another related problem. Transfer learning aims to transfer knowledge from a large dataset known as source domain to a smaller dataset named target domain. In our model, we have used 5674 images of size 100 x 100 x 3 as input. Our input dataset contains images collected from google image search and various video clips from live matches. Some of the techniques used to increase our image dataset are: *Randomized Cropping *Changing contrast in various proportions *Changing brightness *Horizontal flipping
The images are manually annotated and contains two classes: No-ball Legal-ball
We have used Keras and Tensorflow2.0 to build our model and generate results. Our model produces a score for both possible outcomes then each of them is converted to a probability by Sigmoid activation function.
Upload the test data set on google drive. Give the path of the dataset folder on the drive to variable path. Give ‘y’ (correct output of images to be tested) as text file y.txt. The model will print accuracy score, precision, recall and F1 score for the test data
VGG 19
20
30
adam
Accuracy
94.88 %
89.45 %
0.7722
0.9950
0.8696