Sensor-Based Track Adjustment for Low-Latency Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in efficiently updating tracking and identifying of dynamic objects due to delays and inefficiencies in processing new measurement data from multiple sensors.
Innovation Solution
A system and method that utilize a machine learning model to generate adjustments to track data associated with an object, based on input measurement data from sensors and previous track data, allowing for focused update processing without associating new data with the whole scene.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional processing methods are used to update tracking data by associating new measurement data with the whole scene, then comprehensive object detection can be achieved, but processing delays and inefficiencies increase
Solution Approach 1:
The patent segments the tracking update process by dividing measurement data into multiple points and processing each point independently through the machine learning model. Instead of associating all measurement data with the entire scene, the system processes individual measurement points to generate track adjustments, significantly reducing processing complexity and delay while maintaining tracking accuracy.
Solution Approach 2:
The patent extracts only the necessary information for track adjustment from the measurement data using a machine learning model. Rather than processing the complete scene data, the system extracts relevant features from individual measurement points and generates adjustments to existing tracks, eliminating unnecessary processing steps and reducing time loss.
2Reliability
If multiple point radar measurements are processed using conventional methods, then complete object tracking can be maintained, but processing efficiency decreases
Solution Approach 1:
The patent applies segmentation by processing multiple radar measurement points independently through the machine learning model. Each measurement point is handled as a separate input to generate corresponding track adjustments, which maintains complete object tracking information while significantly improving processing efficiency compared to conventional whole-scene association methods.
Solution Approach 2:
The patent replaces the mechanical association process with a machine learning-based adjustment generation system. Instead of using traditional algorithms to associate measurement data with scene objects, the system uses a trained machine learning model to directly generate track adjustments from measurement points, substantially improving processing efficiency while maintaining tracking reliability.
Data Source
AI summary
A method includes obtaining a first track associated with an object. A first set of parameters is generated based on the first track. Measurement data are obtained from one or more sensors. A first set of features are extracted from the measurement data. Based on the first set of parameters and the first set of features, a second set of parameters are generated by a machine learning model. The second set of parameters represent an adjustment to the first set of parameters. Based on the second set of parameters, the first track is adjusted to generate a second track associated with the object. The second track is provided to an autonomous vehicle control system for autonomous control of a vehicle.