Vehicle Motion State Estimation via Feature Point Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle environment detection systems require significant computational effort and bandwidth to estimate the motion state of extended objects, often relying on stereo cameras and motion models, which limits their efficiency and accuracy.
Innovation Solution
A vehicle environment detection system that directly calculates a two-dimensional motion state of extended objects using a linear equation system, tracking feature points with a feature point tracker, and combining velocities to determine a common motion state without model assumptions, utilizing radar and camera sensors for detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If stereo camera and motion model are used for motion state estimation, then measurement precision is improved, but computational effort and bandwidth increase significantly
Solution Approach 1:
The patent extracts only the essential velocity information from feature points directly from image sequences, removing the need for complex stereo camera setups and motion models. By focusing on extracting velocity vectors from optical flow of feature points, the system achieves motion state estimation with reduced computational requirements while maintaining accuracy.
Solution Approach 2:
The patent replaces the mechanical/stereo camera system with a single camera combined with optical flow analysis. Instead of using multiple cameras to capture depth and motion information, the system substitutes this with image processing techniques that extract velocity information directly from sequential images, significantly reducing hardware complexity and computational burden.
2Measurement precision
If velocity estimation from sensor is performed, then measurement precision is improved, but bandwidth usage increases
Solution Approach 1:
The patent creates a virtual copy of velocity information by calculating it from image sequences rather than relying on sensor data. By computing velocity vectors from the optical flow of feature points across frames, the system generates accurate velocity estimates without requiring additional sensor bandwidth, effectively copying motion information from visual data.
3Productivity
If model assumptions are used for motion state calculation, then computational effort is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent enables the system to self-determine motion state by directly calculating velocity vectors from feature point trajectories in image sequences. Instead of relying on pre-defined motion models, the system uses the actual observed motion of feature points to compute velocity information, achieving both computational efficiency and high accuracy through direct measurement.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure relates to a vehicle environment detection system (2) that comprises at least one detection device (3, 4) and at least one processing unit (5), and is arranged to detect at least two feature points (12, 13) at objects (11) outside a vehicle (1). Each feature point constitutes a retrievable point that has a fixed position (x1, y1; x2, y2) on said object (11). The processing unit (5) is arranged to: - determine the positions (x1, y1; x2, y2) and the resulting velocities (vr1, vr2) for each one of said feature points (12, 13) for multiple frames by means of a feature point tracker; and - determine a reference position (x0, y0), a corresponding reference velocity (vr0) and reference angular velocity (ω), constituting a common motion state for the object (11), by means of the results from the feature point tracker; where a feature point tracker is constituted by a tracking algorithm which is arranged to track multiple features and which comprises temporal filtering.