Radar Inverse Sensor Modeling for Occupancy Grid Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Radar sensors face challenges in generating occupancy grids due to indirect reflections, making it difficult to distinguish drivable areas from non-drivable areas, unlike video and lidar sensors which provide clearer reflections along a line of sight.
Innovation Solution
A method and device for generating an inverse sensor model for radar sensors using radar measurement data, obstacle dimensions, and spatial positions, which assigns occupancy probabilities to cells in an occupancy grid, accounting for indirect reflections and using machine learning, specifically convolutional neural networks, to evaluate radar data and integrate with data from other sensors like cameras and lidar.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If radar sensors are used to generate occupancy grids, then obstacle detection capability is improved, but measurement precision deteriorates due to indirect reflections
Solution Approach 1:
The system performs preliminary actions by collecting radar measurement data in advance under controlled conditions with known obstacles at predetermined positions. This training data is used to generate an inverse sensor model before actual obstacle detection, allowing the system to pre-learning the relationship between radar reflections and occupancy states, thereby compensating for indirect reflection effects
Solution Approach 2:
The system implements feedback by using the inverse sensor model to continuously refine occupancy probability assignments. The model processes radar measurement data and provides feedback in the form of occupancy probabilities, which are then used to update the occupancy grid. This closed-loop approach allows the system to iteratively improve measurement precision by leveraging the statistical relationships learned during training
2Measurement precision
If inverse sensor model is generated using training data, then occupancy probability accuracy is improved, but device complexity increases
Solution Approach 1:
The system segments the obstacle detection task into distinct phases: data collection with known obstacles, inverse sensor model generation from training data, and actual occupancy grid generation. By dividing the complex task of real-time obstacle detection with indirect reflections into these manageable segments, the system can pre-compute the inverse sensor model offline, reducing the complexity burden on the real-time detection system
Data Source
Figure 1~2
Figure 3~5
Figure 6~7
AI summary
The invention relates to a method for creating an inverse sensor model for a radar sensor device (21), comprising the following steps: arranging (S1) obstacles (41 to 44) with predefined measurements and spatial positions in an environment of the radar sensor device (21); generating (S2) radar measurement data using the radar sensor device (21); and creating (S3) the inverse sensor model using the generated radar measurement data and the predefined measurements and spatial positions of the obstacles (41 to 44); wherein the inverse sensor model assigns an occupancy probability to a cell (9-ij) of an occupancy grid (9) according to predefined radar measurement data.