Upright Object Detection Using Road Plane Image Flow Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced driver assistance systems (ADAS) and autonomous vehicle (AV) systems face challenges in accurately detecting upright objects due to false positive detections of steep graded roads, which can lead to errors in navigation and control, especially when the road's image flow resembles that of an upright object.
Innovation Solution
A method that utilizes image flow analysis across multiple frames, combined with road plane modeling and validation, to differentiate between road grades and upright objects, incorporating data from various sensors like cameras, radar, and lidar to reduce false positives by segmenting and tracking image patches and validating road plane models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If preconfigured object classifiers are used to detect objects in the environment, then the system can identify predefined objects and control the vehicle or alert the driver, but the system produces false positive detections of steep graded roads as upright objects due to limited ability to handle the variety and dynamic nature of road environments
Solution Approach 1:
The patent segments the detection problem into multiple components: road plane detection, image flow analysis, and upright object detection. By dividing the scene into a road plane portion and analyzing image flow characteristics separately, the system can distinguish between steep graded roads and actual upright objects, reducing false positives without requiring a complete redesign of the detection system
Solution Approach 2:
The patent introduces an intermediary road plane model as a mediator between the camera input and the upright object classifier. This road plane model serves as a reference framework that helps the classifier understand the underlying road geometry, thereby reducing false detections while maintaining the existing classifier structure
2Measurement precision
If the system uses image flow analysis to differentiate between road grades and upright objects, then false positive detections are reduced, but the processing complexity and computational requirements increase
Solution Approach 1:
The patent performs preliminary road plane detection and modeling before conducting upright object detection. By establishing the road plane model first and using it to guide subsequent image flow analysis, the system prepares the detection framework in advance, making the differentiation process more efficient and less computationally intensive
Solution Approach 2:
The patent employs dynamic image flow analysis that adapts to the detected road plane characteristics. The analysis adjusts its parameters and thresholds based on the specific road geometry detected, allowing for precise differentiation without requiring fixed, overly complex processing rules for all possible scenarios
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Systems and methods are provided for augmenting upright object detection. In one implementation, a system for augmenting detection of objects in an environment of a vehicle may include at least one processing device. The at least one processing device may be programmed to: receive, from an image capture device, a first image frame, the first image frame including an attention area associated with a suspected upright object indication; warp, using a level road plane model, an area in a second earlier image frame that corresponds to the attention area; track a plurality of image patches across the warped area and the attention area; compute a road plane model fit to the tracked image patches; and determine whether to suppress the upright object indication based on the tracked image patches being more consistent with a road plane model than with an upright object model.