Object Recognition Integration Device Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition integration devices face inaccuracies in identifying objects due to limitations in detection ranges and reflection intensities of sensors, leading to erroneous exclusions of pedestrians and other objects from detection candidates, even when they are within detection ranges.
Innovation Solution
An object recognition integration device and method that integrate detection data from multiple sensors by determining association relationships between measurement data and previous object data, using certainty values and object types to associate and update object data, thereby improving accuracy in object identification and reducing errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the detection threshold value of laser radar is reduced to detect pedestrians outside the pedestrian detection range, then the detection range is extended, but pedestrians with low reflection intensity are still excluded from detection candidates due to sensor limitations
Solution Approach 1:
The patent combines detection results from multiple sensors (laser radar, millimeter wave radar, optical camera) to create a comprehensive object detection system. By merging the detection capabilities of different sensors with varying strengths, the system extends effective detection range while maintaining reliability, as each sensor compensates for the limitations of others.
Solution Approach 2:
The system implements multi-functional sensor integration where different sensor types (laser radar for range, millimeter wave radar for penetration, optical camera for detailed recognition) work together to achieve universal object detection capabilities across various conditions and distances, ensuring reliable detection regardless of individual sensor limitations.
2Reliability
If the detection threshold value of laser radar is increased to reduce false detections, then detection accuracy improves, but pedestrians within the detection range are missed
Solution Approach 1:
The system employs feedback mechanisms where detection results from multiple sensors are continuously cross-validated. When one sensor detects an object, other sensors provide feedback to confirm or refute the detection, allowing the system to maintain high detection accuracy while avoiding false positives through iterative verification.
Solution Approach 2:
The patent introduces an object recognition integration device as an intermediary that processes and reconciles detection data from multiple sensors. This intermediary component mediates between conflicting detection results, using certainty values and association relationships to determine the most accurate object identification while reducing false detections.
3Productivity
If a detected object is determined as a stationary obstacle based on reflection intensity, then the object is excluded from pedestrian detection candidates, but stationary pedestrians are erroneously excluded as well
Solution Approach 1:
The system applies different quality assessment criteria to different object types and detection contexts. Instead of using a single reflection intensity threshold for all objects, the patent implements localized quality evaluation that considers object position, detection history, and sensor-specific characteristics to accurately distinguish between stationary obstacles and stationary pedestrians.
Solution Approach 2:
The patent dynamically changes detection parameters such as certainty thresholds and association criteria based on the detection context. By adjusting these parameters according to object position, sensor type, and detection confidence levels, the system maintains processing efficiency while improving the accuracy of object type identification and reducing erroneous exclusions.
4Device complexity
If one sensor is selected for detection processing, then the detection process is simplified, but objects not detected by the selected sensor or erroneous identifications cannot be corrected
Solution Approach 1:
The patent segments the detection function across multiple specialized sensors, with each sensor optimized for specific detection tasks. The laser radar handles long-range detection, millimeter wave radar handles penetration through obstacles, and the optical camera handles detailed object recognition. This segmentation allows the system to maintain manageable complexity while achieving high reliability through specialized detection capabilities.
Data Source
AI summary
Provided are an object recognition integration device and an object recognition integration method, which are capable of integrating pieces of detection data that are detected by a respective plurality of sensors in consideration of an inaccuracy of identification of objects. An association relationship between measurement data and previous object data is determined based on an object type and a certainty for each object type candidate contained in measurement data generated for each of the plurality of sensors, and an object type and a certainty for each object type candidate contained in the previous object data. Then, association data is generated by associating the measurement data and the previous object data, which are determined as having “possibility of association”, with each other, to thereby generate current object data by updating the previous object data with use of the association data.