LiDAR Multi-Object Tracking via Hybrid-Time Position Map Averaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking methods require complex association steps and are less versatile due to the need for heuristic matching and hyper-parameters, making them challenging to apply in autonomous driving scenarios.
Innovation Solution
A method for multi-object tracking that generates a hybrid-time position map and a temporary tracked position map based on predicted motion fields and ego-motion, averaging these maps to create a tracked position map without an association step, thereby inheriting object identities and tracking new and occluded objects without additional hyper-parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If tracking-by-detection pipeline with association step is used, then object tracking can be achieved, but processing complexity increases due to complex association steps and hyper-parameters
Solution Approach 1:
The patent extracts and removes the complex association step from the traditional tracking-by-detection pipeline. By using a transformer-based model that directly processes point cloud data to generate tracked position maps, the method eliminates the need for separate detection and association steps, thereby reducing processing complexity while maintaining tracking accuracy
Solution Approach 2:
The transformer-based model serves multiple functions simultaneously: it performs detection, tracking, and association in a unified framework. The model processes point cloud data to generate both object detection results and tracked position maps with identity information, eliminating the need for separate modules and hyper-parameters
2Reliability
If heuristic matching and hyper-parameters are used in association step, then object tracking can be performed, but versatility decreases due to difficulty in applying to different autonomous driving scenarios
Solution Approach 1:
The patent changes the approach from using fixed hyper-parameters to using learnable parameters through the transformer model. The model learns optimal tracking parameters from data, allowing it to adapt to different autonomous driving scenarios without manual tuning, thereby improving versatility while maintaining tracking capability
Solution Approach 2:
The patent replaces the mechanical association step with fixed hyper-parameters with a data-driven transformer model. The model automatically learns and adapts to different scenarios through training, eliminating the need for manual parameter adjustment and improving scenario adaptability
Data Source
AI summary
A method and a device for multi-object tracking, and an electronic device are provided. The method includes: determining a hybrid-time position map of a current point cloud fragment; converting a tracked position map of a previous point cloud fragment into a temporary tracked position map of the current point cloud fragment; and averaging the hybrid-time position map and the temporary tracked position map of the current point cloud fragment, to generate a tracked position map of the current point cloud fragment. With the method and the device for multi-object tracking, and the electronic device, the hybrid-time position map and temporary tracked position map of the current point cloud fragment are averaged, so that not only the tracked position map of the current point cloud fragment is accurately generated, but also an object ID is inherited. Based on the object ID, the same object in different point cloud fragments are associated, so that multi-object tracking is implemented without an association step in the conventional solutions. It is unnecessary to set additional hyper-parameters, and strong versatility is achieved.