Absolute Depth Calculation Using Reference Height in Surveillance Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the coordinates of an element of interest in the real world from images, such as those using neural networks for disparity maps or estimating distances based on bounding box sizes, are inadequate for wide-angle surveillance images due to limitations in calculating absolute depth and are affected by camera positioning.
Innovation Solution
A method that calculates the absolute depth of a reference point in the real world using pixel coordinates and relative depth, allowing for the determination of real coordinates of an element of interest by selecting a remarkable image point with a predefined height and applying calibration parameters like camera inclination and focal distance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If disparity maps are used to estimate depth, then relative depth can be determined, but absolute depth cannot be obtained
Solution Approach 1:
The patent introduces a reference object with known real-world dimensions as an intermediary element. By detecting this reference object in the image and using its known size, the system establishes a scaling factor that converts relative depth measurements from disparity maps into absolute depth values, thereby recovering the lost absolute depth information
Solution Approach 2:
The patent changes the parameter representation by introducing absolute depth calculations based on reference object scaling. Instead of relying solely on relative depth from stereo disparity, the system transforms the depth parameter into absolute measurements by incorporating the known dimensions of the reference object detected in the image
2Ease of operation
If bounding box size is used to estimate distance, then distance can be determined for ground level cameras, but the method fails for surveillance cameras at height
Solution Approach 1:
The patent creates a universal distance estimation method that works for both ground-level and elevated surveillance cameras. By using the reference object's known dimensions and its detected position in the image, the system establishes a camera-position-independent scaling relationship that adapts to any camera height or angle, making the method universally applicable across different surveillance scenarios
3Area of stationary object
If surveillance cameras are used at greater distances, then wide-area coverage is achieved, but depth estimation accuracy deteriorates
Solution Approach 1:
The reference object acts as an intermediary measurement standard that bridges the large distance between the surveillance camera and target objects. By detecting the reference object's known dimensions in the image, the system creates a local scale reference that compensates for the reduced pixel resolution at long distances, thereby maintaining measurement precision across wide surveillance areas
Data Source
Figure 1~2a
Figure 2b~2c
Figure 3a~3b
AI summary
The invention relates to a method for determining, by a computer, from an image (i) taken by a camera (10) comprising an element of interest (E) identified by a plurality of image points (p), the real coordinates (X, Y, Z) of a point of interest (P) in the environment of the camera (10) corresponding to an image point (p) of the plurality of image points (p), the method comprising the following steps: - a step of selecting in the image (i) by the computer a remarkable image point (pr) from among the plurality of image points (p), the remarkable image point (pr) corresponding in the real environment to a remarkable point (Pr) whose order of magnitude of height is known, a predefined height being assigned to the height, - a step of calculating by the computer (20) an absolute depth of the remarkable image point (pr) from a triplet of components (xr, yr,wr) of the remarkable image point (pr) and the predefined height, - a step of determining by the calculator the real coordinates (X, Y, Z) of the point of interest (P) in the real environment of the shooting device, from a triplet of components (x, y, w) of the image point (p) corresponding to the point of interest (P) and the absolute depth.