Camera-Based Cross Traffic Detection for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in urban intersections due to the limitations of radar and Lidar systems, which struggle with detecting and classifying target objects in noisy environments, especially when objects are far away or occluded, and fail to accurately determine velocity for cross-traffic situations.
Innovation Solution
A system and method utilizing cameras to obtain images of the surrounding region, processing them to select context regions, estimate confidence levels, determine bounding boxes, and alter vehicle motion parameters based on target object parameters, including velocity, through feature map extraction and neural network processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Length of stationary object
If radar and Lidar systems are used for target detection, then detection range is extended, but measurement precision deteriorates in urban intersections with noisy background clutter and occluded objects
Solution Approach 1:
The patent introduces camera images as an intermediary information source to supplement radar and Lidar data. The camera provides visual context that helps distinguish target objects from background clutter and occluded objects, thereby improving classification accuracy while maintaining the extended detection range capability of radar and Lidar systems.
Solution Approach 2:
The system fuses multiple sensor types (radar, Lidar, and camera) to create a composite detection system. This multi-sensor fusion approach combines the long-range detection capability of radar/Lidar with the high-precision visual recognition of cameras, resolving the contradiction between detection range and measurement precision.
2Area of stationary object
If radar systems are used for velocity determination, then detection coverage is improved, but velocity measurement accuracy deteriorates for objects moving across the field of view
Solution Approach 1:
The camera serves as an intermediary for velocity measurement by tracking the motion of target objects across successive frames. This visual tracking approach provides accurate velocity information for cross-traffic situations while maintaining the comprehensive detection coverage provided by radar systems.
3Difficulty of detecting and measuring
If Lidar systems are used for target detection, then detection capability is enhanced, but resolution is insufficient for distant target objects
Solution Approach 1:
The camera image acts as an intermediary that provides high-resolution visual information for distant target objects. While Lidar enhances detection capability by detecting objects at various ranges, the camera compensates for the resolution limitation by providing detailed visual features for classification and identification of distant targets.
Data Source
AI summary
A vehicle, system and method of driving of an autonomous vehicle. The vehicle includes a camera for obtaining an image of a surrounding region of the vehicle, an actuation device for controlling a parameter of motion of the vehicle, and a processor. The processor selects a context region within the image, wherein the context region including a detection region therein. The processor further estimates a confidence level indicative of the presence of at least a portion of the target object in the detection region and a bounding box associated with the target object, determines a proposal region from the bounding box when the confidence level is greater than a selected threshold, determines a parameter of the target object within the proposal region, and controls the actuation device to alter a parameter of motion of the vehicle based on the parameter of the target object.