Adaptive Object Detection Region for Radar-Camera Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In vehicle collision avoidance systems using millimeter-wave radar and monocular cameras, objects detected by both systems may be mistakenly recognized as the same when in proximity, even if they are different.
Innovation Solution
An object detection apparatus that defines regions for objects detected by radar and camera, using a learning progress status to estimate the focus of expansion (FOE) in the camera image, allowing accurate determination of object overlap in the XY-plane, thereby distinguishing between radar and camera-detected objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If objects detected by radar and camera are compared using fixed region definitions, then the determination process is simple, but false positives increase when objects are in proximity
Solution Approach 1:
The patent applies dynamics by making the camera detection region adaptive rather than fixed. The region definition changes based on the detected object's distance from the own vehicle, with the depthwise length being adjusted according to distance. This dynamic adjustment allows the system to maintain high identification accuracy while avoiding false positives, as the region size automatically adapts to the spatial relationship between the own vehicle and detected objects
Solution Approach 2:
The patent implements parameter changes by modifying the depthwise length parameter of the camera detection region based on the distance to the detected object. When objects are farther away, the depthwise length is increased; when objects are closer, the depthwise length is decreased. This parameter adjustment resolves the contradiction by allowing the system to maintain reliable object identification without the complexity of multiple fixed region definitions
2Quantity of substance
If the depthwise length of camera detection region is increased to cover more area, then more objects are detected, but false positives increase with radar objects
Solution Approach 1:
The system uses dynamic region adjustment where the depthwise length of the camera detection region is automatically adjusted based on the distance to the detected object. This allows the region to expand when objects are far away (detecting more objects) and contract when objects are close (reducing false positives with radar), thereby resolving the contradiction between detecting more objects and maintaining identification accuracy
Solution Approach 2:
The patent changes the depthwise length parameter of the camera detection region according to the distance parameter. By making this parameter variable rather than fixed, the system can detect a appropriate number of objects at each distance while maintaining reliable identification, preventing both missed detections and false positives
Data Source
AI summary
In an object detection apparatus, a first region definition unit defines a first object region including a first detection point representing a relative position of a first object detected based on detection information from the radar. A second region definition unit defines a second object region including a second detection point representing a relative position of a second object detected based on a captured image from a monocular camera. A learning status acquisition unit acquires a learning progress status to estimate a position of FOE on the captured image. If there is an overlap of the first and second object regions, a determination unit determines that these objects are the same object. The second region definition unit sets a length of the second object region in a depthwise or vehicle-lengthwise direction representing a direction of the second detection point with respect to the reference point depending on the learning progress status.