The man was fined 380 euros for scratching his head while driving
A Dutch man was fined 380 euros after an AI camera caught him "talking on the phone while driving." However, he claims that he was actually scratching his temple at the time and the system made a mistake. SSPDaily tells about it.
This was reported by SSPDaily, which published the corresponding footage.
Last November, Tim Hansen received a fine for allegedly talking on his cell phone while driving a month earlier. He was shocked, mainly because he didn't remember using his phone while driving that day, so he decided to check the incriminating photo with the Central Collection Agency.
At first glance, it looks like Tim is actually talking on the phone, but a closer look reveals that he's not actually holding anything in his hand. He was just scratching his temple, and the camera took the position of his hand as if he were holding a smartphone. What's even more puzzling is that the person who checked the photo and confirmed the fine also didn't notice the "false positive."
Hansen, who works in information technology creating algorithms for image editing and analysis, used his personal experience to explain how the Monocam police camera system works and why it can make mistakes. Although he was unable to test Monocam himself, he explained how the system works and why it can give false positives.
"If the model is supposed to predict whether something is a yes or a no, of course it can happen that the model is wrong," Tim said. "In my case, the model indicated that I was holding the phone, but I wasn't. Then we talk about a false positive. An ideal model would only predict true positives and true negatives, but a 100% correct prediction is rare."
The IT specialist explained that systems like Monocam need to be trained on a large set of images divided into two or three groups: a training set, a validation set, and a test set. The first set is used to teach the algorithm which objects are in which images and what properties (colors, lines, etc.) belong to them, the second to optimize a number of algorithmic hyperparameters, and the third to test how well the system actually works.
"The algorithm we used, as well as the police algorithm, may suspect the presence of a phone because the training dataset contains many examples of people making a phone call with their hand next to their ear," said Tim. "It is possible that the training dataset contains few or no photos of people sitting with an empty hand over their ear. In this case, it becomes less important to the algorithm whether the phone is actually in the hand, but it is enough that the hand is close to the ear. To improve the situation, more photos should be added where the hand is empty."
Due to the many variables that can affect the algorithm's decision, a human filter is needed to minimize the number of false positives, Hansen says.
The driver has appealed the fine and expects a positive outcome, but now he will have to wait up to 26 weeks for an official verdict. His case has gone viral in the Netherlands and neighboring countries such as Belgium, where some institutions are asking for cameras capable of detecting cell phone use while driving, but Tim's story proves that they are far from 100% reliable.