This week, Tesla was granted a patent for a system that utilizes image data from a vehicle’s camera along with a trained machine learning model to identify the distance of an object from a vehicle.
The need for this particular technology arises due to the increasing cost and complexity of vision sensors for autonomous driving systems in mass market vehicles.
The background within Tesla’s patent also states, “each additional sensor increases the input bandwidth requirements for the autonomous driving system. Therefore, there exists a need to find the optimal configuration of sensors on a vehicle. The configuration should limit the total number of sensors without limiting the amount and type of data captured to accurately describe the surrounding environment and safely control the vehicle.”
About the Patent Application
Tesla’s invention enables a method of using neural networks to determine the distance from a vehicle to an object using only image data. There are two neural networks in this system: one capable of determining distance to objects purely from image data, and a second one to create training material for the first neural network.
The first neural network receives images captured using the camera of a vehicle and identifies objects within the image to extract an estimate of the distance to the identified object.
The second neural network correlates distance sensor data to objects within images captured by the vehicle to create annotated images. These annotated images are used by the first neural network as training material to further improve its capability to extract the distance to an object using only image data. As the patent describes, “the training data can be automatically generated and used to train a machine learning model to predict object properties with a high degree of accuracy.”
WHAT THE INVENTION WOULD DO
This method enables a vehicle to detect and interpret the distance to its surroundings using image data. Through the use of camera image data, the system essentially adds a second pair of eyes to the vehicle. There are many safety features that can be newly enabled or further improved with this type of image analysis, as well as improving on self-driving vehicle systems like Tesla’s Autopilot.
why This Patent Application is Interesting
Traditionally processes are controlled with some sort of feedback loop based on the current output of the system. Through the use of isolated neural networks, more specifically an executing neural network and a training neural network, the system is able to continuously improve on itself regardless of the quality of the instantaneous feedback. These types of neural are changing the way that we analyze current processes and demonstrates furtherance of both control systems and control theory.
Written by John DeStefano, Technical Advisor
and Lauren Hawksworth, Marketing Administrator
March 24, 2021