Depth-Based Object Verification for False Image Detections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis methods for autonomous vehicles and robots are prone to false object detections, leading to unexpected maneuvers that can cause accidents, as visual cues may misleadingly indicate the presence of objects that are not actually present.
Innovation Solution
A method that verifies the presence of objects in images using depth information from sensors like monocular estimation, stereo cameras, or electromagnetic interrogation, determining if expected depth changes align with detected objects, employing operators like the Sobel filter and morphological closing to enhance depth representation, and validating object detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual cues are used for object detection in images, then object detection capability is improved, but false detection rate increases
Solution Approach 1:
The patent introduces depth information as an intermediary verification layer between visual detection and final object confirmation. The depth map derived from stereo vision or monocular estimation serves as a mediator that validates whether detected objects have corresponding geometric depth characteristics, thereby reducing false detections while preserving true object detection capability
Solution Approach 2:
The patent transitions from two-dimensional image plane analysis to three-dimensional space verification by incorporating depth information. By mapping detected objects onto a depth map and verifying their presence in the third dimension, the system adds a spatial verification dimension that eliminates false positives from 2D visual cues alone
2Measurement precision
If depth information is obtained from multiple sensors, then verification accuracy is improved, but device complexity increases
Solution Approach 1:
The patent employs sensors that serve multiple functions: the same sensor system used for primary object detection in the image is also utilized for depth estimation. This multi-functionality approach allows verification without adding dedicated separate sensors, thereby improving accuracy while controlling device complexity
Solution Approach 2:
The patent creates a virtual depth model (depth map) that copies and represents the three-dimensional structure from two-dimensional image data through computational algorithms. This copying approach enables depth verification without requiring physical depth sensors, reducing hardware complexity while maintaining verification capability
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Reduces false object detections by confirming the presence of objects through geometric cues, ensuring accurate object recognition and reducing inappropriate system responses.
Implementation Method 1
depth information obtained by measuring a distance to the scenery using a beam of electromagnetic interrogation radiation
Data Source
AI summary
A method for verifying the presence of objects detected in a scenery based on at least one image of this scenery. The method includes: assigning, by a given object detector, at least one image region in the image to an object whose presence in the scenery the image region indicates; obtaining depth information relating to this image region, the depth information being indicative of a distance between a sensor that was used to acquire the image and a scenery region in the scenery that corresponds to the image region; determining whether the depth information includes depth changes that are to be expected given the presence of the object assigned to the image region by the object detector; and if this determination is positive, determining that the assignment of the image region to the object by the object detector is a valid detection of the object.