Inverse Radar Grid Mapping for Static-Dynamic Object Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Evidential grid mapping algorithms struggle to distinguish artifacts due to noise in radar sensor data from dynamic objects, and do not adequately account for sensor characteristics like antenna gain patterns, affecting the accuracy of occupancy grid maps.
Innovation Solution
Implement inverse radar sensor modeling and grid mapping techniques that utilize motion and radar inputs to determine the probability of cells being occupied by stationary, dynamic, or free objects, incorporating sensor characteristics to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If evidential grid mapping algorithms process radar sensor data to create occupancy grid maps, then the environment representation is generated, but artifacts from noise and dynamic objects cannot be distinguished, reducing map quality
Solution Approach 1:
The patent segments the occupancy probability into multiple discrete states (free, static, dynamic, static-dynamic) rather than treating it as a continuous probability. This segmentation allows the system to distinguish between different types of occupied cells, enabling differentiation between noise artifacts and actual dynamic objects through state classification.
Solution Approach 2:
The patent introduces an intermediary classification layer that processes radar measurements before updating the occupancy grid. This intermediary layer evaluates measurements against multiple criteria (velocity thresholds, consecutive detection requirements, state transition rules) to determine whether a measurement represents noise, a static object, or a dynamic object, thereby mediating between raw sensor data and final map representation.
2Device complexity
If traditional grid mapping algorithms are used, then processing is simpler, but sensor characteristics like antenna gain patterns are not accounted for, affecting probability accuracy
Solution Approach 1:
The patent performs preliminary action by pre-characterizing the radar sensor's antenna gain pattern and storing it as a lookup table or function before grid mapping operations. This pre-computed sensor characteristic is then applied during occupancy probability calculations to correct for directional sensitivity variations, improving accuracy without adding computational complexity during the main mapping process.
3Productivity
If radar sensor measurements are processed without motion compensation, then processing is faster, but accuracy is reduced when the radar is mounted on a moving platform
Solution Approach 1:
The patent applies motion compensation by calculating the radar platform's motion vector and using it to counterweight or correct the velocity measurements of detected objects. By subtracting the platform's motion contribution from the raw velocity measurements, the system can more accurately determine whether detected velocity comes from static objects (should show no relative velocity after compensation) or truly dynamic objects.
Data Source
AI summary
An example device includes motion sensing and processing circuitry to generate compensated motion data for a first time based on raw motion data and a set of motion indicators including a velocity indicator for the device calculated based on the raw motion data; a radar sensor to receive reflections indicating detections and generate data points for the first time representing the detections, in which each data point includes position and velocity information of a corresponding detection relative to the radar sensor; a first circuit to generate object data for the first time based on the set of the data points and the compensated motion data for the first time; and a second circuit to calculate, based on the object data for the first time and a characteristic of the radar sensor, for each cell in a grid representing an FOV of the radar sensor at the first time, probabilities of the cell being in a free state, a stationary state, and a dynamic state.