Vehicle Trajectory Tracking via Object Detection Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle collision avoidance systems rely on GPS data, which is susceptible to error and unavailable in certain environments, such as tunnels or areas with dense tree cover, necessitating an alternative method for accurate vehicle trajectory tracking.
Innovation Solution
A method using object detection data from sensors like radar and cameras to calculate the vehicle's pose by identifying point and extended objects, combining this data with vehicle dynamics to determine the actual trajectory and compare it to a planned path for collision avoidance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS data is used for vehicle trajectory tracking, then the system can provide a second source of trajectory data, but the system becomes dependent on GPS signal availability and cannot operate in tunnels or areas with dense tree cover
Solution Approach 1:
The patent introduces an intermediary system (sensor-based object detection and pose estimation system) that mediates between the vehicle dynamics data and the trajectory tracking function. This intermediary uses sensors to detect objects and calculate vehicle pose relative to these objects, providing trajectory information that works independently of GPS signal availability.
Solution Approach 2:
The patent replaces the GPS-based mechanical/systematic approach with a sensor-based optical/electromagnetic approach. Instead of relying on GPS satellites and radio signals, the system uses sensors to detect objects and calculate pose through image processing and geometric relationships, substituting the GPS mechanism with a different physical principle.
2Speed
If vehicle dynamics sensors are used for calculating actual vehicle trajectory, then the system can provide real-time trajectory data, but cumulative error accumulates over time
Solution Approach 1:
The patent implements feedback by continuously detecting objects in the environment and using their known positions to correct and update the vehicle pose estimate. The system compares detected object positions with expected positions based on the vehicle dynamics model, and uses this feedback to correct cumulative errors in real-time trajectory tracking.
Solution Approach 2:
The patent creates a copy of the environment by detecting and tracking objects in the scene. By maintaining a model of object positions and using these detected object positions as reference points, the system can accurately determine vehicle pose without accumulating error, effectively copying environmental features for navigation purposes.
3Measurement precision
If object detection data and iterative pose calculation are used to determine vehicle trajectory, then the system achieves high measurement precision without cumulative error, but the device complexity increases
Solution Approach 1:
The patent makes the sensor system multi-functional by using the same sensors for both object detection and vehicle pose estimation. The object detection data serves dual purposes: identifying environmental obstacles and providing reference points for calculating vehicle position and orientation, thereby reducing the need for separate specialized hardware.
Solution Approach 2:
The patent segments the trajectory tracking function into distinct modules: object detection, pose estimation, and trajectory calculation. This segmentation allows each module to be optimized independently and makes the complex system more manageable and implementable using existing sensor technologies and processing algorithms.
Data Source
AI summary
A method for determining an actual trajectory of a vehicle using object detection data and vehicle dynamics data. An object detection system identifies point objects and extended objects in proximity to the vehicle, where the point objects are less than a meter in length and width. An updated vehicle pose is calculated which optimally transposes the point objects in the scan data to a target list of previously-identified point objects. The updated vehicle pose is further refined by iteratively calculating a pose which optimally transposes the extended objects in the scan data to a target model of previously-identified extended objects, where the iteration is used to simultaneously determine a probability coefficient relating the scan data to the target model. The updated vehicle pose is used to identify the actual trajectory of the vehicle, which is compared to a planned path in a collision avoidance system.