Cascaded Object Tracking with Affinity Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for tracking objects in autonomous vehicles process multiple cues independently, leading to unreliable object tracking due to lack of collaboration between cues.
Innovation Solution
A cascaded tracking method that calculates affinity scores between currently detected objects and previously tracked objects using multiple cues, allowing for the modification of affinity scores in a progressive manner to associate objects and determine whether to skip affinity score calculations based on previously calculated scores, thereby improving tracking accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple cues are processed independently in conventional methods, then the processing is simple and fast, but the object tracking reliability is poor
Solution Approach 1:
The patent combines multiple independent cue processing streams into a unified tracking framework where cues collaborate through affinity score modification. The affinity matrix integrates information from multiple cues (appearance, motion, depth) to jointly determine object tracks, resolving the contradiction by merging separate processing paths into a coordinated system that improves reliability while maintaining manageable complexity through structured integration.
2Measurement precision
If all affinity scores are calculated for all object pairs using multiple cues, then the tracking accuracy is maximized, but the computational time and resources increase significantly
Solution Approach 1:
The patent performs preliminary affinity score calculations using a first cue to establish initial matches before proceeding to subsequent cue processing. By pre-calculating affinity scores with the first cue and using them to guide subsequent processing, the system identifies obvious matches early, allowing it to skip or reduce calculations for certain object pairs in later stages, thus maintaining high accuracy while reducing overall computational time.
Solution Approach 2:
The patent implements a skipping mechanism where, after calculating affinity scores with the first cue, the system determines whether to skip affinity score calculations for certain object pairs in subsequent cue processing. If the first cue already provides sufficient discrimination or if the computational cost outweighs the benefit, the system skips redundant calculations, significantly reducing processing time while preserving tracking accuracy for critical object pairs.
3Productivity
If multiple cues are processed in a cascaded manner with progressive affinity score modification, then the tracking reliability is improved and unnecessary calculations are reduced, but the system complexity increases
Solution Approach 1:
The patent segments the cue processing into distinct stages (first cue processing, subsequent cue processing) with clear separation of functions. Each stage handles specific aspects of affinity score calculation and modification, allowing the complex multi-cue problem to be divided into manageable segments that can be processed sequentially. This segmentation improves processing efficiency by organizing computations in a structured pipeline while keeping each segment's complexity manageable through clear functional boundaries.
Data Source
AI summary
In one embodiment, objects are tracked by calculating affinity scores with multiple cues in cascading stages. In each stage, affinity scores are calculated for currently detected objects based on a cue. The affinity scores are then fed into subsequent stages to inform subsequent calculations based on subsequent cues, where calculations can be skipped and resources saved. Object tracks are updated every frame, to assist in autonomous vehicle driving.