Moving Object Detection Using Average Scene Disparity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision systems for detecting moving objects from a moving platform under low illumination conditions face challenges in accurately identifying objects of any shape without relying on computationally expensive ego-motion calculations and are prone to failure in filtering processes due to the complexity of enumerating all possible human shapes.
Innovation Solution
A computer vision system utilizing an imaging subsystem with infrared-sensitive cameras to capture scene image data, a computing subsystem for calculating average scene disparity to differentiate between stationary and moving objects, and an output subsystem to provide motion detection signals, which bypasses explicit ego-motion computation by using a simplified formula and a two-stage correspondence matching approach for stereo image processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional computer vision algorithms are used to detect moving objects from a moving platform, then object detection capability is achieved, but computational complexity and processing time increase due to explicit ego-motion computation
Solution Approach 1:
The patent extracts and removes the computationally expensive explicit ego-motion computation step from the conventional computer vision pipeline. Instead of calculating ego-motion separately and then compensating for it, the invention directly computes object motion relative to the camera using simplified geometric relationships, thereby eliminating the harmful computational burden while preserving detection accuracy.
Solution Approach 2:
The patent inverts the conventional approach by not trying to determine camera motion first and then subtracting it from observed motion. Instead, it directly computes the motion of objects relative to the camera frame by analyzing the geometric constraints of the scene, effectively working backwards from the observed image plane motion to infer object motion without explicit ego-motion calculation.
2Reliability
If conventional pedestrian detection methods are used under low illumination conditions, then human detection is achieved, but the filtering process becomes complicated and prone to failure due to difficulty in enumerating all possible human shapes
Solution Approach 1:
The patent changes the detection parameters from shape-based features (which require enumerating all possible human shapes) to motion-based features (displacement vectors and velocity). By detecting objects that move independently of the background flow, the system reliably identifies pedestrians without needing to model complex human shape variations, thereby simplifying the filtering process while improving reliability.
Solution Approach 2:
The patent replaces the mechanical shape-matching system with a field-based motion analysis system. Instead of mechanically comparing image features against predefined human shape templates, the system uses field-wide optical flow analysis to detect motion patterns characteristic of pedestrians, substituting a complex mechanical recognition process with a more robust field-based approach.
3Illumination intensity
If infrared-sensitive cameras are used for night vision, then detection capability under low illumination is improved, but the appearance of the scene becomes different from daytime requiring separate processing algorithms
Solution Approach 1:
The patent creates a universal motion detection algorithm that works across different illumination conditions and imaging modalities. By base the detection on fundamental geometric relationships and motion patterns rather than appearance-specific features, the same algorithm effectively processes both daytime visible light images and nighttime infrared images without requiring separate processing pipelines.
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 efficient detection of moving objects under low illumination conditions without requiring complex ego-motion calculations, improving the accuracy and speed of pedestrian detection in vehicular applications by classifying objects based on estimated and actual position shifts, thereby enhancing automotive safety.
Implementation Method 1
an imaging subsystem with infrared-sensitive cameras to capture scene image data
Data Source
AI summary
Moving object detection is based on low illumination image data that includes distance or depth information. The vision system operates on a platform with a dominant translational motion and with a small amount of rotational motion. Detection of moving objects whose motions are not consistent with the movement of the background is complementary to shape-based approaches. For low illumination computer-based vision assistance a two-stage technique is used for simultaneous and subsequent frame blob correspondence. Using average scene disparity, motion is detected without explicit ego-motion calculation. These techniques make use of characteristics of infrared sensitive video data, in which heat emitting objects appear as hotspots.