Monocular Camera Object Detection via Top-Down View Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting obstacles using monocular cameras are unreliable due to assumptions about static scenes and require high computational costs, especially when dealing with dynamic environments.
Innovation Solution
The method involves obtaining and transforming image data from successive frames to a predetermined point of view, extracting and matching features, computing angular changes, and detecting objects based on consistent angular changes, which allows for reliable object detection even in dynamic scenes using a monocular camera.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If stereo cameras or ultrasound detection means are used for reliable three-dimensional obstacle detection, then detection reliability is improved, but device complexity and cost increase
Solution Approach 1:
The patent uses a monocular camera to capture images and creates a virtual top-down view by transforming the captured images through coordinate transformations. This virtual copy of the scene from a different perspective enables reliable object detection without requiring complex stereo camera hardware or ultrasound sensors, thus maintaining detection reliability while reducing device complexity and cost.
2Measurement precision
If structure from motion is used for estimating three-dimensional properties, then depth estimation capability is improved, but computational cost increases
Solution Approach 1:
The patent transforms the coordinate system of captured images to create a top-down view representation, changing the parameter space from standard camera coordinates to a transformed coordinate system. This parameter transformation enables depth and distance estimation through geometric relationships in the transformed view, achieving accurate depth measurement with reduced computational requirements compared to full structure from motion algorithms.
3Ease of manufacture
If static scene assumptions are made for object detection, then detection simplicity is improved, but reliability deteriorates in dynamic environments
Solution Approach 1:
The patent captures multiple images at different time points and transforms them to a common top-down view coordinate system. By comparing features across these transformed images and analyzing angular changes, the system can dynamically adapt to moving objects and changing scenes, maintaining detection reliability in dynamic environments while keeping the detection approach relatively simple.
Data Source
AI summary
An enhanced object detecting method and apparatus is presented. A plurality of successive frames is captured by a monocular camera and the image data of the captured frames are transformed with respect to a predetermined point of view. For instance, the images may be transformed in order to obtain a top-down view. Particular features such as lines are extracted from the transformed image data, and corresponding features of successive frames are matched. An angular change of corresponding features is determined and boundaries of an object are identified based on the angular change of the features.