Multi-Vehicle Sensor Fusion for Long-Range Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensors in driver-less cars, such as radar and lidar, face challenges in accurately recognizing the shape of objects, particularly at distances beyond their effective range, leading to reduced recognition accuracy.
Innovation Solution
A method involving a neural network that synchronizes sensing information from multiple sensors on a moving object with interworking moving objects, using one sensor's data as ground truth to enhance recognition accuracy by training and correcting the other sensor's data, specifically using radar and lidar sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If lidar sensor is used to recognize objects three-dimensionally, then measurement precision is improved, but reliability deteriorates due to sensitivity to external environment
Solution Approach 1:
The patent combines data from multiple sensors (lidar, radar, camera) to create a more reliable recognition system. By fusing information from different sensing modalities, the system maintains measurement precision while compensating for individual sensor vulnerabilities to environmental factors.
Solution Approach 2:
The patent introduces an intermediary processing system that mediates between raw sensor data and final object recognition. This intermediary layer processes and validates data from multiple sources, reducing the direct impact of environmental interference on recognition reliability.
2Reliability
If radar sensor is used to detect objects, then reliability is improved, but measurement precision deteriorates due to inaccurate shape recognition
Solution Approach 1:
The patent merges radar detection data with lidar and camera data to compensate for radar's shape recognition limitations. The combined system maintains radar's reliable detection capability while adding precise shape information from other sensors.
Solution Approach 2:
The patent creates a multi-functional sensing system where each sensor type contributes its strengths: radar provides reliable detection, lidar provides precise 3D measurement, and camera provides detailed visual information. The system universally handles both detection and precise characterization.
3Measurement precision
If neural network training with synchronized multi-sensor data is implemented, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent performs preliminary synchronization and preprocessing of multi-sensor data before neural network training. By preparing synchronized datasets in advance with proper time alignment and coordinate transformation, the system reduces the complexity of real-time processing while maintaining high measurement precision.
Solution Approach 2:
The system implements self-service mechanisms where the neural network automatically learns to handle the complexity of multi-sensor fusion during training. The network adapts to synchronize and integrate data from multiple sources, reducing the need for complex external synchronization hardware or software.
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
Improves the recognition accuracy of objects beyond the effective range of individual sensors by leveraging synchronized data from interworking moving objects, enhancing the neural network's estimation performance.
Implementation Method 1
training a neural network for estimating recognition information of the target from the first synchronized sensing information by using information on the target included in the second synchronized sensing information as ground truth data
Data Source
AI summary
A method and device for object recognition for information collected from a sensor of a moving object are disclosed. The method may include identifying a target with a confidence level of recognition accuracy less than a threshold, based on first sensing information collected from a moving object, obtaining second sensing information on the target from an interworking moving object, synchronizing the first sensing information and the second sensing information based on time information included in the first sensing information and time information included in the second sensing information, and training a neural network for estimating recognition information of the target from the first synchronized sensing information by using information on the target included in the second synchronized sensing information as ground truth data.