Vehicle Environment Pattern Recognition With Spatial-Temporal Attention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning algorithms for autonomous driving and ADAS struggle to explicitly correlate sequentially recorded data to differentiate between distinct objects and incorporate legacy information for improved object tracking and velocity estimation.
Innovation Solution
A computer-implemented method using an attention algorithm to generate a joint spatial-temporal data set by correlating current and memory data, enabling explicit pattern recognition and tracking of object dynamics through grid maps, with training via loss functions to optimize neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If machine learning algorithms including recurrent units are applied to sequential sensor scans, then spatial and temporal patterns can be derived, but the model cannot explicitly differentiate between distinct objects or incorporate legacy information for improved tracking
Solution Approach 1:
The patent segments the processing of sequential data by introducing discrete object hypotheses with unique identifiers. Each object is tracked independently through separate feature clusters in the memory state, allowing explicit differentiation between distinct objects while maintaining their individual dynamic properties across time steps.
Solution Approach 2:
The patent introduces an intermediary object hypothesis layer that mediates between raw sensor data and pattern recognition. This intermediary structure with unique identifiers and associated feature clusters enables the model to explicitly track and differentiate objects while incorporating legacy information through the persistent hypothesis structure across sequential time steps.
2Duration of action of moving object
If recurrent units are used to aggregate sensor data over time, then a memory state is generated, but the model cannot explicitly correlate sequential data to improve velocity estimation and object matching
Solution Approach 1:
The patent implements feedback mechanisms where the memory state containing legacy object dynamics information is fed back into the current processing step. This feedback loop allows the model to explicitly correlate sequential data, using past velocity and position information to improve current velocity estimation and object matching accuracy through iterative refinement.
Solution Approach 2:
The patent performs preliminary action by pre-establishing object hypotheses with unique identifiers and associated feature clusters before final pattern recognition. This preliminary structuring of temporal data with explicit object identities enables more accurate velocity estimation and tracking by organizing legacy information in a ready-to-use format for correlation with current sensor data.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method is provided for determining patterns related to an environment of a host vehicle from sequentially recorded data. Respective sets of characteristics detected by a perception system in the environment of the host vehicle are determined for a current point in time and for a predefined number of previous points in time. A set of current input data associated with the sets of characteristics is generated for the current point in time, and a set of memory data is generated by aggregating the sets of characteristics for the previous points in time. An attention algorithm is applied to the set of current input data and to the set of memory data to generate a joined spatial-temporal data set, and at least one pattern is determined for the environment of the host vehicle from the joined spatial-temporal data set.