Approaching Object Detection via Segmented Optical Flow and Dynamic Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaching object detection systems face challenges in accurately detecting moving objects with high computation load, particularly for vehicles, and suffer from erroneous detection due to high processing requirements and low accuracy in vertical edge detection.
Innovation Solution
The system performs optical flow processing to detect velocity vectors in a specific direction, using dynamic programming to determine object movement, thereby reducing computation load and improving detection accuracy by processing image portions in a different direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical flow processing is performed on all pixels in all directions to detect moving objects accurately, then detection accuracy is improved, but computation load increases significantly
Solution Approach 1:
The patent segments the computation process into two distinct stages: first, optical flow processing is performed only in the horizontal direction to extract moving regions; second, dynamic programming is applied only to the vertical direction for size change detection. This segmentation of computational tasks across different spatial directions reduces the overall computation load while maintaining detection accuracy.
Solution Approach 2:
The patent transitions from two-dimensional optical flow computation (both horizontal and vertical directions) to a separated approach where horizontal motion detection and vertical size change detection are performed as independent one-dimensional processes. This dimensional separation allows each process to operate more efficiently with reduced computational requirements.
2Speed
If optical flow processing is performed in all directions for real-time detection, then detection speed is improved, but computation load becomes too high for car-mounted microcomputers
Solution Approach 1:
The detection process is segmented into two sequential steps performed by different detection units: horizontal moving region detection via optical flow, followed by vertical size change detection via dynamic programming. This segmentation enables real-time processing by distributing computational tasks across multiple specialized units rather than requiring one microcomputer to handle all computations simultaneously.
Solution Approach 2:
The patent applies partial action by performing optical flow processing only in the horizontal direction initially, and dynamic programming only in the vertical direction, rather than performing full two-dimensional optical flow. This partial computation approach achieves sufficient detection performance with significantly reduced computation load suitable for car-mounted systems.
3Measurement precision
If vertical edge detection is performed to identify moving object regions, then region identification accuracy is improved, but erroneous detection occurs due to low accuracy in vertical edge detection
Solution Approach 1:
Instead of detecting vertical edges directly (which has low accuracy and causes erroneous detection), the patent inverts the approach: it first detects horizontal motion via optical flow to identify moving regions, then uses dynamic programming to detect vertical size changes. This inverted detection sequence avoids the problematic vertical edge detection step while achieving reliable moving object identification.
Solution Approach 2:
The patent introduces dynamic programming as an intermediary process between horizontal motion detection and final approaching object determination. This intermediary step processes the horizontally-detected moving regions through vertical size change analysis, providing a more reliable pathway to identify approaching objects without relying on inaccurate vertical edge detection.
Data Source
AI summary
An approaching object detection system, approaching object can be accurately detected while reducing the load on a calculation processing. A first moving region detection unit (30) detects (#3) only an optical flow substantially in horizontal direction in an image (P) captured by a camera (10) (#1) and sets (#4) a rectangular image portion (Q) containing a region moving in a substantially horizontal direction according to the detected optical flow. A second moving region detection unit (40) obtains a distribution (profile) of signal values of the set rectangular image portion (Q), (#5) in a vertical direction (longitudinal direction). Furthermore, the second moving region detection unit (40) correlates distributions of signal values of image portions (Q, Q) of two continuous images (P, P) in time series (#6) based on dynamic programming based processing to obtain an enlargement rate (#7). An approaching object determining unit (50) determines according to the enlargement rate whether the object is actually approaching.


