Single-Image Pose Estimation From Motion Blur in Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face high computational complexity and errors in localization due to the need for processing multiple images, and obstacles like rain or snow can hinder proper localization using conventional methods.
Innovation Solution
A machine learning model processes single images from sensors to determine changes in vehicle pose based on blurring, reducing computational complexity and enhancing localization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple images are processed for localization, then localization accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent extracts and utilizes motion blur information from individual images as a separate useful signal. Instead of requiring multiple images to determine motion, the system isolates the blur component caused by vehicle movement and uses it directly for pose estimation, thereby reducing computational complexity while maintaining localization accuracy
Solution Approach 2:
The patent replaces the mechanical/computational system of capturing and processing multiple sequential images with an optical/algorithmic approach that processes a single image. By substituting the multi-image capture mechanism with motion blur analysis of one image, the system reduces processing requirements while achieving the same localization function
2Measurement precision
If multiple images are processed for localization, then pose estimation is improved, but processing speed decreases
Solution Approach 1:
The system extracts motion blur characteristics from a single image to directly compute pose changes. By isolating the blur component as the primary indicator of motion, the system eliminates the need to process multiple images sequentially, thereby significantly increasing processing speed while maintaining accurate pose estimation
Solution Approach 2:
The patent skips the intermediate step of capturing and processing multiple sequential images by directly analyzing motion blur in a single image. This rushing through the localization process using blur information from one frame rather than multiple frames enables real-time processing at higher speeds
3Reliability
If conventional multi-image localization is used, then localization can be performed, but reliability decreases when sensors are blocked
Solution Approach 1:
The system uses the motion blur inherently present in the single captured image to perform localization. Since the blur is caused by the vehicle's own motion during the exposure period, the system serves itself by extracting localization information from its own motion signature in the image, making it reliable even when external objects or weather block the sensor view
Data Source
AI summary
A vehicle comprises a sensor configured to capture images and one or more processors. The one or more processors can be configured to receive a single image from the sensor, the single image captured by the sensor as the autonomous vehicle was moving; execute a machine learning model using the single image as input to generate a change in pose of the autonomous vehicle, the machine learning model trained to output changes in pose of autonomous vehicles based on blurring in individual images; determine a global position of the autonomous vehicle based on the generated change in pose of the autonomous vehicle; and transmit the global position to an autonomous vehicle controller configured to control the autonomous vehicle.