Mastering Computer Vision with TensorFlow 2.x

Training the cascade detector

The overall training system was developed to maximize the detection rate and minimize the false-positive rate. Viola and Jones achieved this by setting the target detection rate and false-positive rate for each stage of the cascade detector as follows:

• The number of features in each layer of the cascade detector is increased until the target detection rate and the false-positive targets are achieved for that layer.
• Another stage is added if the overall false-positive rate is not low enough.
• A rectangular feature is added to the current stage until its target rates have been met.
• The false-positive target from the current stage is used as a negative training set for the next stage.

The following image shows ...