Camera Object Validation via Homography Plane Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle detection systems based on image classification often misclassify road structures as vehicles due to their local focus, leading to unnecessary warnings or interventions in driver assistance systems.
Innovation Solution
A method that records a sequence of images from a vehicle camera, detects potential collision objects, determines corresponding features, calculates homographies to assign features to planes in space, and validates objects by minimizing back projection errors to distinguish between road structures and raised objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If classification-based systems are used for object detection, then detection speed and simplicity are improved, but false detection and misclassification increase due to road structures resembling vehicles
Solution Approach 1:
The patent transitions from 2D image classification to 3D spatial reasoning by calculating homographies and assigning features to planes in space. This dimensional elevation allows the system to distinguish between road structures (which lie on the ground plane) and raised objects (vehicles, pedestrians) that protrude from it, thereby maintaining detection speed while significantly improving accuracy.
Solution Approach 2:
The patent introduces homography calculation and plane assignment as intermediary steps between image capture and object classification. These intermediaries transform raw image features into spatially contextualized data, enabling the system to filter out false detections from road structures before final classification occurs.
2Device complexity
If local environment analysis is used for object classification, then processing complexity is reduced, but false detection increases due to inability to distinguish road structures from vehicles
Solution Approach 1:
The patent segments the image space into distinct planes (ground plane, raised objects) through homography calculation. By dividing the scene into spatially separated planes, the system can independently analyze each plane's characteristics, reducing processing complexity while improving the accuracy of object validation through plane-specific feature analysis.
Solution Approach 2:
The patent adds a spatial plane dimension to the classification process. Instead of merely analyzing local image patches, the system elevates features to 3D space, assigns them to planes, and uses plane properties (such as whether an object protrudes from the ground plane) to validate detections, thereby improving accuracy without proportionally increasing complexity.
3Reliability
If more validation steps are added to distinguish road structures from objects, then detection accuracy is improved, but processing time and computational load increase
Solution Approach 1:
The patent performs preliminary plane assignment and homography calculation on all detected features before final object validation. By pre-organizing features into spatial planes, the system eliminates the need for complex post-detection validation, reducing overall processing time while maintaining high accuracy through structured spatial reasoning.
Data Source
Figure 1~2
Figure 3~4
Figure 5~5d
AI summary
The invention relates to a method and to a device for detecting and validating objects from images from a camera and can be used in particular in camera-based driver assistance systems. The method for detecting objects from a series of images from a vehicle camera comprises the following steps: a) capturing a series of images by means of the vehicle camera, b) detecting a potential collision object in at least one image, c) determining corresponding features in two consecutive images in the region of the detected potential collision object, e) associating a plurality of determined corresponding features with a plane in space, and g) validating a potential collision object if the associated plane corresponds to a projecting object. According to a preferred embodiment, the method comprises the following step(s): d) computing homographs for the determined corresponding features for associating said corresponding features with a plane in space; and/or e2) associating a plurality of determined corresponding features with one of a plurality of planes of specified orientation in space, and f) making an association with the plane in space that results in the smallest back-projection error for the determined corresponding features, wherein the back-projection error indicates the difference between the measured correspondence of a feature in two consecutive images and the correspondence predicted from the computed homograph.