Multi-Stage Object Recognition for Automated Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition systems in automated driving face challenges in accurately recognizing objects at varying distances due to the differing spatial and temporal resolutions of imaging devices and ranging sensors.
Innovation Solution
The system employs a multi-stage object recognition process that adapts to short, middle, and long-distance regions by utilizing measurement data from a ranging sensor and image data from a camera, with specific processes optimized for each region to leverage the strengths of both sensor types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single object recognition process is used for all distance ranges, then the system complexity is reduced, but the object recognition accuracy varies across different distance regions
Solution Approach 1:
The patent divides the measurement range into multiple distance regions (first distance region for short distances, second distance region for medium distances, third distance region for long distances). Each region has its own optimized object recognition process that leverages the strengths of specific sensor types for that distance range, thereby improving overall recognition accuracy without requiring a single complex universal process
Solution Approach 2:
The patent dynamically selects and adjusts the object recognition process based on the distance to the target object. The system transitions between different recognition methods (ranging sensor-based, camera-based, or fused) depending on the current distance region, making the system adaptable to varying operational conditions rather than using a fixed single process
2Speed
If only ranging sensor data is used for object recognition, then the processing speed is fast for short distances, but the recognition accuracy deteriorates for middle and long distances
Solution Approach 1:
The patent segments the distance range and assigns different primary sensors to different segments: ranging sensor for the first distance region (fast processing), camera for the second distance region (better accuracy), and fused data for the third distance region (longest range). This segmentation allows each sensor to operate in its optimal performance zone
Solution Approach 2:
The patent uses data fusion as an intermediary approach for the third distance region, combining ranging sensor data and camera data to achieve accurate object recognition at long distances where neither sensor alone would be sufficient, thus bridging the gap between speed and accuracy requirements
3Measurement precision
If only camera image data is used for object recognition, then the recognition accuracy is good for middle distances, but the processing becomes complex and slow for short distances
Solution Approach 1:
The patent segments the operational range so that camera-based recognition is primarily used for the second distance region where it provides good accuracy without excessive complexity. For the first distance region, the system switches to simpler ranging sensor-based recognition, avoiding unnecessary computational complexity
Solution Approach 2:
The patent changes the operational parameters (which sensor or fusion method to use) based on the distance parameter. By adjusting the selected recognition process according to distance, the system optimizes the balance between processing complexity and accuracy for each specific operational condition
4Measurement precision
If data fusion is performed for all distance regions, then the object recognition accuracy is maximized, but the processing time and computational load increase
Solution Approach 1:
The patent segments the distance range and applies data fusion only where necessary (third distance region for long distances). For closer distances (first and second regions), the system uses simpler single-sensor recognition methods, thereby reducing processing time and computational load when full fusion is not needed
Solution Approach 2:
The patent applies data fusion partially rather than universally - only for the third distance region where it provides the most benefit. This partial application of the fusion technique avoids the excessive processing time and computational load that would result from applying fusion to all distance regions
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances object recognition accuracy across different distance regions by effectively utilizing the unique characteristics of imaging devices and ranging sensors, thereby improving the overall performance of automated driving systems.
Implementation Method 1
a ranging sensor, configured to measure at least distances to reflection points based on reflected waves of applied irradiation waves
Data Source
AI summary
An object recognition apparatus performs an object recognition process based on measurement data of a ranging sensor and image data of a camera. The object recognition apparatus: performs a first object recognition process for a short-distance region, a second object recognition process for a middle-distance region, and a third object recognition process for a long-distance region; performs the object recognition process using the measurement data in the first object recognition process; performs the object recognition process using the measurement data and the image data in the second object recognition process; and adjusts, based on data on a traveling environment of a mobile body, respective time periods for which the first object recognition process and the second object recognition process are to be performed, or a ratio between respective ranges in which the first object recognition process and the second object recognition process are to be performed in a measurement range.