Object Tracking via Bayesian Inference and LIDAR
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle systems face challenges in accurately determining and navigating around objects in their environment, particularly in dynamic traffic scenarios, due to limitations in sensor data processing and object tracking.
Innovation Solution
The implementation of a computing device with LIDAR sensors and Bayesian inference algorithms to classify and track objects in real-time, forming estimated enclosing rectangles to improve object detection and vehicle navigation, enabling safe autonomous operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional sensor data processing methods are used for object tracking, then the system complexity remains low, but the object detection accuracy and tracking reliability deteriorate in dynamic traffic scenarios
Solution Approach 1:
The patent segments the object tracking problem into distinct phases: detection phase (identifying potential objects in sensor data) and tracking phase (following objects across multiple time steps). Each phase has specialized processing logic, with detection focusing on initial object identification and tracking focusing on continuity and refinement. This segmentation improves detection accuracy while managing system complexity by avoiding monolithic processing.
Solution Approach 2:
The system performs preliminary actions by establishing initial object detections before formal tracking begins. These preliminary detections serve as starting points for subsequent tracking operations, allowing the system to prepare tracking structures in advance and improve overall tracking accuracy without adding significant computational burden during critical real-time operation.
2Reliability
If simple object tracking methods are used, then the processing speed remains high, but the reliability of object tracking deteriorates in complex traffic environments
Solution Approach 1:
The patent implements dynamic tracking where object parameters (position, velocity, dimensions) are continuously updated based on new sensor measurements and previous track states. The tracking model adapts to changing traffic conditions by adjusting object predictions in real-time, improving reliability in complex environments while maintaining processing speed through efficient recursive updates rather than complete re-detection.
Solution Approach 2:
The system employs feedback mechanisms where tracking results from previous time steps inform current detection and tracking decisions. The predicted object states from tracking are compared with new sensor measurements, and discrepancies are used to refine object estimates. This feedback loop improves tracking reliability by continuously correcting predictions based on actual observations while maintaining computational efficiency through incremental updates.
3Reliability
If detailed object classification is performed, then the navigation safety improves, but the computational time and processing load increase
Solution Approach 1:
The patent applies local quality by performing detailed object classification and analysis selectively based on object relevance to the ego-vehicle. Objects closer to the vehicle or in critical paths receive more detailed processing and classification, while distant or irrelevant objects use simpler tracking. This approach improves navigation safety for critical objects while reducing overall computational time by avoiding uniform detailed processing of all detected objects.
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
Enhances the accuracy and reliability of object detection and tracking, allowing vehicles to navigate safely and efficiently in complex traffic environments by iteratively refining object estimates based on sensor data and prior observations.
Implementation Method 1
a first LIDAR sensor of the plurality of LIDAR sensors
Implementation Method 2
determining a first object based on a first plurality of object three-dimensional (3D) data points
Data Source
AI summary
A computing device in a vehicle can determine an object by determining a first rectangular bound centered on the object based on joint probabilities of first 3D data points, then determining the object by determining a second rectangular bound centered on the object based on joint Bayesian probabilities of second 3D data points conditioned on the first rectangular bound and piloting the vehicle based on the determined object.