Camera Object Pose Classification via Precomputed Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting and identifying randomly oriented objects in a digital camera's field of view are slow, making them less effective for real-time applications and less desirable for robotic systems.
Innovation Solution
A method that generates classifier feature vectors for various facet viewing angles of an object, compares image feature vectors with these classifiers, and determines the object's pose based on the closest match, using a digital camera and image classifier to rapidly classify objects in a field of view.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If known methods for detecting and identifying randomly oriented objects are used, then object identification can be achieved, but the response time is extended and processing speed is slow
Solution Approach 1:
The patent segments the object recognition process by creating multiple classifier feature vectors, each corresponding to a specific facet viewing angle. Instead of processing the entire object at once, the system divides the recognition task into discrete angular segments, allowing for faster comparison and identification of the object's pose by matching the captured image against these pre-segmented classifiers.
Solution Approach 2:
The patent applies preliminary action by pre-computing and storing multiple classifier feature vectors for different facet viewing angles before the actual object recognition occurs. This preparation work is done in advance, so that during real-time operation, the system only needs to compare the captured image against these pre-prepared classifiers, significantly reducing processing time.
2Productivity
If digital camera is mounted on robotic arm for real-time object recognition, then robotic interaction is improved, but the slow response time lessens efficacy
Solution Approach 1:
The system performs preliminary computations by pre-generating classifier feature vectors for all possible facet viewing angles before robotic operation. This allows the robotic system to quickly identify objects in real-time without performing complex computations during operation, thereby improving robotic interaction efficiency while maintaining fast response times.
Solution Approach 2:
The patent changes the parameter representation by transforming object recognition into a parameter-matching problem where the captured image parameters are compared against pre-stored classifier parameters for different viewing angles. This parameter transformation enables faster processing and improves the efficacy of real-time robotic interaction.
Data Source
AI summary
A method for classifying a known object in a field of view of a digital camera includes developing a plurality of classifier feature vectors, each classifier feature vector associated with one of a plurality of facet viewing angles of the known object. The digital camera captures an image in a field of view including the known object and an image feature vector is generated based upon said captured image. The image feature vector is compared with each of the plurality of classifier feature vectors and one of the plurality of classifier feature vectors that most closely corresponds to the image feature vector is selected. A pose of the known object relative to the digital camera is determined based upon the selected classifier feature vector.