Vehicle Sensing System for Moving Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for detecting moving objects, particularly approaching or retreating objects, face challenges in accuracy and computational load, especially in complex motion scenarios like blind spot monitoring for motorcycles, where background and object motion are intertwined, leading to errors and increased processing demands.
Innovation Solution
A sensing system that directly detects moving objects by computing 2D motion vectors between consecutive images and classifying points based on transformation parameters, eliminating the need for background motion estimation, using sparse optical flow techniques and triangulation methods to reduce computational effort and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If background motion detection is performed first to enable moving object detection, then moving objects can be detected by comparing their motion with background motion, but errors in background detection propagate to all subsequent processing steps and increase computational load
Solution Approach 1:
The patent extracts and analyzes local motion patterns directly from the image sequence without performing global background motion detection first. By computing optical flow vectors and analyzing their local transformation properties (scaling, rotation, translation) directly at each point, the system avoids propagating errors from a global background model while still enabling accurate moving object detection through local motion pattern analysis
Solution Approach 2:
The patent segments the image into local regions and analyzes motion patterns independently in each region. By dividing the image into patches and computing optical flow and transformation parameters locally for each patch, the system avoids the need for a single global background model, thereby preventing error propagation while maintaining detection accuracy across the entire scene
2Measurement precision
If tracking-based estimation with Kalman filter is used to calculate motion vectors, then motion vectors can be predicted over time, but the system requires proper state initialization, prediction, and has higher detection delay
Solution Approach 1:
The patent performs preliminary computation of optical flow vectors and transformation parameters for all image points in each frame before performing moving object detection. By pre-computing these motion characteristics and organizing them into connectivity graphs and interest point tracks in advance, the system eliminates the need for iterative Kalman filter prediction during detection, thereby reducing detection delay while maintaining estimation accuracy
Solution Approach 2:
The patent skips the iterative prediction and state initialization steps of Kalman filtering by directly computing transformation parameters from optical flow vectors and connectivity graph analysis. This approach rushes through the computation by using direct algebraic methods rather than iterative optimization, significantly reducing detection delay while achieving comparable or superior accuracy through local transformation analysis
3Loss of information
If dense motion estimation is used to describe movement of each individual image pixel, then complete motion information is obtained, but computational load increases significantly
Solution Approach 1:
The patent applies local quality by computing dense optical flow and transformation parameters only in regions where motion patterns are significant or change rapidly. By analyzing local transformation properties (scaling, rotation, translation) for each patch and selectively processing regions with high motion variance or interest points, the system maintains complete motion information where needed while reducing computational load in static or uniformly moving regions
Data Source
AI summary
The present invention presents a sensing system 1 and a corresponding method for detecting moving objects 11 in the surroundings of a vehicle 10. The sensing system 1 comprises an imaging unit 2 for obtaining an image stream 2a, a computing unit 3 for analyzing the image stream 2a, and a control unit 4 for controlling the vehicle 10 based on the analysis result of the computing unit 3. The sensing system 1 particularly employs a background-model-free estimation. The sensing system 1 is configured to perform a local analysis of two neighboring motion vectors 6, which are computed from points 7a in images 5a, 5b of the image stream 2a, and to determine, whether the points 7a corresponding to these motion vectors 6 belong to a particularly moving object.