Object Counting via Optical Flow and Partial Image Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object counting methods, such as video surveillance systems, struggle to accurately count objects in densely populated areas where objects overlap or are blocked, leading to erroneous evaluations and inability to determine direction of movement.
Innovation Solution
A device with a counting module that increments the object counter by a partial value for each image where an object moves in the defined counting direction, using surveillance cameras to extract moving image regions and calculate optical flow, allowing for directional counting and error correction through statistical means.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If video surveillance systems track moving objects for a certain period to count them, then the count represents the number of moving objects, but erroneous evaluations occur when objects overlap or are situated too closely
Solution Approach 1:
The patent segments the counting process into discrete image-by-image evaluations rather than continuous tracking. Each image is independently analyzed for moving objects, and counts are accumulated across the image sequence. This segmentation approach avoids the reliability issues of continuous tracking when objects overlap, as each frame is evaluated independently for object presence and motion.
Solution Approach 2:
The patent applies partial counting by evaluating only the moving portions of objects in each image frame rather than requiring complete object tracking. By counting objects that appear in any part of the counting section across multiple frames, the system achieves more reliable counts in crowded scenes where complete object delineation is difficult.
2Measurement precision
If light barriers are used to count moving objects, then individual objects can be detected, but the light barrier can be blocked by stationary objects
Solution Approach 1:
The patent replaces the mechanical light barrier system with a video-based optical flow analysis system. Instead of using physical light barriers that can be blocked, the system uses sequential image capture and computational analysis of motion patterns. This substitution eliminates the blockage problem entirely, as the system analyzes pixel-level motion in video frames rather than relying on light transmission through a fixed barrier.
3Ease of operation
If contactless inductance measurement is used to count cars, then cars can be counted without contact, but the method cannot determine the direction of the moving object
Solution Approach 1:
The patent introduces dynamic analysis through optical flow computation, which captures the temporal evolution of pixel intensities across video frames. By analyzing the direction and magnitude of pixel displacement between consecutive frames, the system determines both the presence and direction of moving objects. This dynamic approach preserves directional information while maintaining contactless detection capabilities.
Solution Approach 2:
The patent introduces optical flow fields as an intermediary between the raw video data and the final counting result. The optical flow computation serves as a mediator that extracts motion direction information from the video sequence, enabling the system to determine object direction without direct contact or additional sensors.
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 reliable and stable object counting in crowded scenes by evaluating image data on an image-by-image basis, reducing errors from overlapping objects and providing a statistical statement on the number of counted objects.
Implementation Method 1
at least one surveillance camera for recording the surveillance region, wherein the surveillance camera is designed to output an image data stream of the surveillance region
Implementation Method 2
an extraction module designed to extract moving image regions in a current image from the image data flow
Data Source
AI summary
For operators of public or commercial buildings, such as supermarkets, it is interesting to determine how many persons, in particular potential buyers, enter the building each day. In other areas, as well, entering and exiting moving objects are counted: it is typical, for example, that cars driving into and out of a parking garage are counted. The invention relates to a device 1 for counting objects in a monitoring area, having a counting module 9, wherein the counting module comprises an object counter increasing by one counting value when one of the objects completely passes through a counting section 6 in a counting direction 4 as a counting object, having at least one monitoring camera 2 for capturing the monitoring area, wherein the monitoring camera 2 is designed for outputting an image data stream of the monitoring area, and having an extraction module 10 designed for extracting moving image areas in a current image of the image data flow, wherein the moving image areas can represent the objects or partial segments thereof, wherein the counting module 9 is designed for increasing the object counter 9 by a partial value for each current image for one of the moving image areas 3, the direction of motion thereof corresponding to the counting direction 4 and thus representing a potential counting object or a partial segment thereof, said partial value being less than the counting value.