|
1. A method comprising: capturing a plurality of images of drivers of motor vehicles through windows of the motor vehicles; selecting from the captured images, a plurality of positive images which show the drivers exhibiting a behavior within the motor vehicles that constitutes a traffic violation; selecting from the captured images, a plurality of negative images which show the drivers not exhibiting the behavior while driving the motor vehicles; selecting a plurality of training images from both the positive images and the negative images; inputting to a machine the selected training images and respective labels indicating that the selected training images are positive images or negative images to train the machine to classify that images, captured through the windows of the motor vehicles, show the drivers exhibiting the behavior while driving the motor vehicle; testing the trained machine by first inputting to the trained machine a positive test image captured through a window of a motor vehicle, wherein the positive test image shows a driver exhibiting the behavior while driving the motor vehicle; responsive to the first inputting, first outputting a first score indicative of confidence of positive identification that the positive test image shows the driver exhibiting the behavior while driving the motor vehicle; masking a feature of the positive test image, thereby producing a masked positive test image with a masked feature; and second outputting a second score indicative of confidence of positive identification that the masked positive test image shows the driver exhibiting the behavior while driving the motor vehicle; wherein a comparison of the first and the second scores indicates relevance of the masked feature to positive identification that the positive test image shows a driver exhibiting the behavior while driving the motor vehicle.
|
