System And Method Of Predicting Human Interaction With Vehicles

What is this patent about?

’046 is related to the field of data analytics, specifically predicting human behavior in relation to vehicles. The background involves the challenge of enabling autonomous vehicles to accurately predict the actions of pedestrians, cyclists, and other drivers, a crucial ability for safe navigation, especially in urban environments. Current autonomous systems often rely solely on motion vectors, which are insufficient for anticipating behavior based on contextual understanding.

The underlying idea behind ’046 is to leverage human intuition to train a predictive model. This is achieved by presenting human observers with images or video clips of road scenes and collecting their judgments about the likely actions or intentions of individuals in those scenes. These judgments are then aggregated into statistical data that serves as the training data for a machine learning model. The trained model can then be used to predict the behavior of road users in real-time scenarios.

The claims of ’046 focus on a computer-implemented method, a non-transitory computer readable storage medium, and a computing system that all perform the same core steps. These steps involve storing images of road users, generating training data by soliciting and aggregating human judgments about the state of mind of those users, training a model based on this data, and then using the trained model to predict the state of mind of users in new images.

In practice, the system captures images or video from a vehicle's sensors. These images are then presented to human observers, potentially through crowdsourcing platforms. The observers are asked to assess the intentions or likely actions of individuals in the images, such as whether a pedestrian intends to cross the street. The responses are collected and used to generate summary statistics, such as the average likelihood of a particular action. These statistics, along with the original images, are used to train a machine learning model, such as a deep convolutional neural network .

The trained model can then be deployed in a vehicle to analyze real-time sensor data and predict the behavior of road users. This allows the vehicle to anticipate potential hazards and make more informed decisions. Unlike prior approaches that rely solely on motion vectors, this system incorporates human intuition and contextual understanding, leading to more accurate and reliable predictions. The system can also be refined by comparing its predictions with actual outcomes and adjusting the model's weights accordingly, creating a feedback loop for continuous improvement.