Object Bounding Contours Using Image and Lidar Depth Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems rely on bounding boxes for object detection, which fail to accurately represent the size and shape of objects, especially those with protrusions or irregular shapes, leading to inaccurate navigation and potential safety hazards.
Innovation Solution
Generating bounding contours using image and lidar data to create more accurate geometric shapes that follow the outer surface of objects, including irregularities, providing a tighter fit and better representation of objects for navigation and collision avoidance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bounding boxes are used for object detection, then the detection process is simple and fast, but the accuracy of representing object size and shape deteriorates
Solution Approach 1:
The patent segments the object detection process into multiple stages: initial bounding box detection, contour extraction from image data, and refinement using depth information. This segmentation allows the system to maintain the speed of bounding box methods while improving accuracy through subsequent processing steps that extract precise contours and incorporate depth data to represent complex object geometries.
Solution Approach 2:
The patent transitions from two-dimensional bounding boxes to three-dimensional bounding volumes by incorporating depth information from depth sensors or stereo vision. This dimensional enhancement allows the system to accurately represent objects with protrusions and irregular shapes by adding the depth dimension, thereby improving measurement precision without excessive complexity increase.
2Measurement precision
If bounding boxes are used for object detection, then the processing speed is high, but the navigation accuracy deteriorates
Solution Approach 1:
The patent performs preliminary bounding box detection to quickly identify objects of interest, then applies more computationally intensive contour extraction and depth-based refinement only to these detected objects. This preliminary action approach maintains high processing speed for the overall scene while achieving high navigation accuracy for specific objects that require precise measurement.
Solution Approach 2:
The patent applies different levels of processing quality to different objects based on their relevance to navigation. Critical objects near the vehicle's path receive full contour extraction and depth-based refinement for high precision, while distant or less relevant objects use simpler bounding box representations. This local quality differentiation optimizes the balance between processing time and navigation accuracy.
3Reliability
If bounding boxes are used for object detection, then the system complexity is low, but the collision avoidance effectiveness deteriorates
Solution Approach 1:
The patent introduces contour extraction as an intermediary step between bounding box detection and collision avoidance processing. The extracted contours provide more accurate object boundaries and shapes, serving as a better intermediary representation for collision risk assessment. This intermediary layer improves collision avoidance effectiveness by providing more precise geometric information without requiring complete system redesign.
Solution Approach 2:
The patent combines multiple data sources (image data, depth data, and bounding box information) to create a composite object representation. This composite approach integrates the simplicity of bounding boxes with the precision of contour extraction and depth information, achieving high collision avoidance effectiveness by leveraging the strengths of multiple detection methods in a unified system.
Data Source
AI summary
Techniques are discussed herein for controlling autonomous vehicles within a driving environment, including generating and using bounding contours associated with objects detected in the environment. Image data may be captured and analyzed to identify and/or classify objects within the environment. Image-based and/or lidar-based techniques may be used to determine depth data associated with the objects, and a bounding contour may be determined based on the object boundaries and associated depth data. An autonomous vehicle may use the bounding contours of objects within the environment to classify the objects, predict the positions, poses, and trajectories of the objects, and determine trajectories and perform other vehicle control actions while safely navigating the environment.