Single Camera Object Velocity Determination Using Image Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for autonomous vehicles to determine the velocity of objects in their environment are often inaccurate, especially at distances where lidar is less effective and in directions where radar is less accurate, requiring multiple cameras and complex image analysis, which increases processing resources and reduces efficiency.
Innovation Solution
A technique that uses a single camera to determine object velocity by transforming a sequence of images into a transformed space, applying a machine-learned model to calculate relative object velocity, and then scaling it based on depth measurements to obtain actual velocity, without the need for multiple cameras or extensive processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple cameras and complex image analysis are used to determine object velocity, then measurement precision may be improved, but device complexity and processing resources increase
Solution Approach 1:
The patent extracts and utilizes only the essential information needed for velocity calculation from the image sequence, specifically focusing on optical flow patterns and motion vectors rather than processing complete image data. This selective extraction approach enables accurate velocity measurement using a single camera while reducing processing complexity.
Solution Approach 2:
The patent replaces complex mechanical multi-camera systems with a computational approach using a single camera. By substituting physical complexity with algorithmic processing of image sequences, the system achieves comparable or superior velocity measurement accuracy with reduced hardware complexity.
2Measurement precision
If multiple cameras and complex image analysis are used to determine object velocity, then measurement precision may be improved, but processing resources and time increase
Solution Approach 1:
The patent segments the image processing task into discrete computational steps: detecting feature points, calculating optical flow between frames, determining motion vectors, and computing velocity. This segmentation enables efficient processing by handling each step independently and optimizing each operation, thereby improving productivity while maintaining precision.
Solution Approach 2:
The patent applies partial action by processing only the necessary portions of image data required for velocity calculation, such as tracking specific feature points and motion vectors rather than analyzing entire images. This selective processing reduces computational load and improves processing efficiency while maintaining measurement accuracy.
3Measurement precision
If lidar is used for velocity determination, then measurement precision may be improved at close range, but effectiveness decreases at distances where lidar is less effective
Solution Approach 1:
The patent creates a universal velocity determination method using a single camera that works effectively across multiple distances and environmental conditions. By employing computational optics and image sequence analysis, the system achieves adaptability across different operational scenarios where lidar performance degrades, making the system universally applicable.
4Measurement precision
If radar is used for velocity determination, then measurement precision may be improved, but accuracy decreases in certain directions where radar is less accurate
Solution Approach 1:
The patent uses image sequences and computational algorithms as an intermediary between the camera and velocity measurement. This intermediary processing layer extracts motion information from visual data, providing accurate velocity measurements in all directions without the directional limitations inherent in radar systems.
Data Source
AI summary
Techniques are discussed for determining a velocity of an object in an environment from a sequence of images (e.g., two or more). A first image of the sequence is transformed to align the object with an image center. Additional images in the sequence are transformed by the same amount to form a sequence of transformed images. Such sequence is input into a machine learned model trained to output a scaled velocity of the object (a relative object velocity (ROV)) according to the transformed coordinate system. The ROV is then converted to the camera coordinate system by applying an inverse of the transformation. Using a depth associated with the object and the ROV of the object in the camera coordinate frame, an actual velocity of the object in the environment is determined relative to the camera.