Camera Range Sensor Fusion for Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current range sensors in vehicles have limited resolution in lateral measurements and often confuse objects due to their small aperture size and inability to accurately track kinematic characteristics of vehicles in adjacent lanes, while cameras are not reliable for longitudinal range measurements due to assumptions about flat ground which is rarely true.
Innovation Solution
A system integrating a camera and range sensor with an on-board computer that fuses data using prediction algorithms to estimate object location and appearance, comparing predicted tracks with existing database tracks for accurate motion analysis and updating radar and camera registration parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a range sensor with small aperture size is used, then the device complexity is reduced and cost is lowered, but the lateral measurement resolution deteriorates and objects in adjacent lanes cannot be accurately tracked
Solution Approach 1:
The patent combines a range sensor (radar) and a camera into a fused sensor system. The range sensor provides accurate longitudinal distance measurements while the camera provides high-resolution lateral position information. By merging these two sensors with complementary strengths, the system achieves both low device complexity and high measurement precision without requiring a single complex sensor.
2Device complexity
If a camera with limited field-of-view is used, then the device complexity is reduced, but the ability to detect objects in blind-spot areas deteriorates
Solution Approach 1:
The camera system is designed to serve multiple functions: it provides detailed lateral position information for objects in the main field-of-view, detects objects in blind-spot areas, and contributes to overall scene understanding. The camera's data is integrated with range sensor data to create a comprehensive environmental model that covers both central and peripheral areas.
3Device complexity
If ground flatness is assumed for range estimation, then the calculation complexity is reduced, but the accuracy of longitudinal range measurement deteriorates on uneven terrain
Solution Approach 1:
The system uses the camera as an intermediary to detect the bottom part of objects adjacent to the ground. By visually identifying the ground contact point and using perspective geometry, the system can estimate range without assuming flat ground. The camera image serves as a mediator between the sensor and the ground, allowing accurate range estimation on uneven terrain.
4Productivity
If only one or two points per object are tracked by the range sensor, then the processing speed is improved, but the ability to distinguish close objects deteriorates
Solution Approach 1:
The system merges the range sensor's longitudinal tracking data with the camera's lateral position data. While the range sensor tracks one or two points per object for speed, the camera provides additional lateral position information that enables distinction between closely spaced objects. The fused data creates a more complete object representation that maintains processing speed while improving object distinction accuracy.
Data Source
AI summary
A transportation vehicle configured to track an object external to the vehicle. The vehicle includes a camera, a range sensor, and an on-board computer. The on-board computer includes a processor and a tangible, non-transitory, computer-readable medium comprising instructions that, when executed by the processor, cause the processor to perform select steps. The steps include determining that new-object data corresponding to the object is available based on input received from the sensor sub-system of the vehicle. The steps also include registering the new-object data and estimating an expected location and an expected appearance for the object according to a prediction algorithm to generate a predicted track corresponding to the object. The steps also include analyzing motion for the object including comparing the predicted track with any existing track associated with the object and stored in a database of the on-board computer.