Sports Video Motion Direction Detection and Target Tracking Algorithm Based on Convolutional Neural Network
Table 1
Training network based on bootstrapping algorithm.
1. Randomly selected 50 positive and 50 negative samples from the initial training to develop the test procedure. It will be used to select the recommended weight in grades 3 and 8
2. Set momentum parameter and false alarm threshold
3. Iterate the training network 50 times, and use the same number of positive and negative samples in each iteration. Set
4. From 100 video frames, collect samples whose detection false-positive results exceed ThrFa, and collect up to 3000 new samples
5. Add the newly collected samples to the negative sample training set
