Multi-Angle View Synthesis from Pre-Registered Image Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image-based navigation systems require additional image collection and prior specification of the object-of-interest for creating multi-angle views, making them inflexible and time-consuming, and unable to efficiently handle occlusions or virtual objects.
Innovation Solution
The system generates multi-angle views from existing image datasets by registering images with GPS and range-finding data, allowing users to specify any object-of-interest and virtually navigate through the dataset, with optional occlusion removal and virtual object addition, without the need for additional image collection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If additional images are collected to create multi-angle views of an object-of-interest, then the quality and completeness of the views improve, but the time and resources required increase significantly
Solution Approach 1:
The system performs preliminary action by collecting a comprehensive image dataset covering the entire locality before any object-of-interest is specified. This pre-collected dataset includes images from multiple locations and angles, which can then be rapidly processed to generate multi-angle views of any object without requiring additional image collection. The preliminary image collection eliminates the need for time-consuming on-site filming when a user wants to view an object from multiple angles.
Solution Approach 2:
The system achieves universality by creating a single comprehensive image dataset that serves multiple purposes: it can generate multi-angle views of any object-of-interest regardless of which object is selected, support virtual navigation through the locality, and provide views from any virtual location. This multi-functional dataset eliminates the need for separate image collection campaigns for each object or viewing scenario.
2Manufacturing precision
If a multi-angle video is created for a specific object-of-interest, then detailed views of that object are available, but the system becomes inflexible and cannot accommodate different objects without recreating the video
Solution Approach 1:
The system implements universality by building a comprehensive image dataset that can serve any object-of-interest within the captured locality. The dataset includes images from multiple locations, orientations, and distances, allowing the system to generate high-quality multi-angle views of any object without requiring video recreation. Users can freely select different objects of interest, and the system will automatically generate appropriate views from the pre-collected dataset.
Solution Approach 2:
The system uses copying by creating virtual representations of objects from the image dataset rather than requiring physical re-filming. When a user selects an object-of-interest, the system synthesizes multi-angle views by selecting and processing appropriate images from the dataset, effectively creating a virtual copy of the object from multiple perspectives without needing to physically recreate the filming scenario.
3Ease of operation
If images are registered with GPS and range-finding data, then the ability to navigate and select views improves, but the processing complexity increases
Solution Approach 1:
The system performs preliminary registration of images with GPS coordinates and range-finding data during the initial image collection phase. This pre-processing establishes the spatial context for all images in the dataset, enabling efficient virtual navigation and view selection later without requiring complex real-time processing. The spatial metadata is attached to each image during collection, so that when users navigate virtually, the system can quickly identify and retrieve appropriate images based on the user's virtual position and orientation.
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
Enables flexible and efficient creation of multi-angle views for any object-of-interest, including virtual ones, without additional image collection, and enhances visibility by removing occlusions, allowing users to navigate through space and time.
Implementation Method 1
The images can be 'registered,' that is, when the images are collected, the objects in the images are associated with their geographic locations. For example, a GPS unit on the camera tells where the camera was located when an image was taken
Implementation Method 2
A range-finding device (e.g., LIDAR) tells how far away an object in the image is from the location of the camera
Data Source
AI summary
Disclosed are a system and method for creating multi-angle views of an object-of-interest from images stored in a dataset. A user specifies the location of an object-of-interest. As the user virtually navigates through the locality represented by the image dataset, his current virtual position is determined. Using the user's virtual position and the location of the object-of-interest, images in the image dataset are selected and interpolated or stitched together, if necessary, to present to the user a view from his current virtual position looking toward the object-of-interest. The object-of-interest remains in the view no matter where the user virtually travels. From the same image dataset, another user can select a different object-of-interest and virtually navigate in a similar manner, with his own object-of-interest always in view. The object-of-interest also can be “virtual,” added by computer-animation techniques to the image dataset. For some image datasets, the user can virtually navigate through time as well as through space.