Multi-Object Tracking via Quantum Annealing Optimization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing multiple object tracking (MOT) methods face scalability issues due to exponential computational growth with the number of objects and frames, making them impractical for large datasets.

Innovation Solution

A computer-implemented method using an annealing solver to optimize binary quadratic optimization problems, mapping the assignment problem into an unconstrained binary quadratic optimization (QUBO) format, which is more efficiently solvable, and leveraging quantum adiabatic computation for scalable solutions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If discrete optimization algorithms are used for multi-object tracking assignments, then long-term relations can be included and transparency is improved, but computational complexity grows exponentially with the number of objects and frames

Engineering Contradiction:
Improvetracking accuracyVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent transforms the discrete optimization problem into a continuous quadratic programming problem by relaxing the binary assignment constraints. This parameter change allows the use of efficient convex optimization algorithms instead of exponential discrete optimization, resolving the contradiction between tracking accuracy and computational complexity.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces auxiliary variables and constraints that can be easily computed and discarded during the optimization process. These temporary computational objects enable the transformation to a more efficient problem formulation without requiring expensive computational resources.

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

2Quantity of substance

If the number of frames and objects to be tracked increases, then the problem size and tracking coverage are improved, but discrete optimization methods become infeasible

Engineering Contradiction:
Improvenumber of tracked objectsVSAvoidcomputational efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

By changing the problem parameters from discrete binary variables to continuous variables with quadratic constraints, the patent enables scalable solutions that can handle large numbers of objects and frames efficiently, resolving the infeasibility of discrete optimization for large-scale tracking.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces the mechanical discrete optimization process with a mathematical continuous optimization framework. This substitution allows the use of efficient numerical solvers that scale polynomially rather than exponentially, enabling handling of large tracking problems.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If machine learning methods are used for assignments, then computational efficiency is improved, but transparency and interpretability are reduced

Engineering Contradiction:
Improvecomputational efficiencyVSAvoidtransparency
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent replaces opaque machine learning models with a transparent quadratic programming formulation. The objective function and constraints explicitly encode the assignment logic, maintaining interpretability while achieving computational efficiency through continuous optimization rather than discrete methods.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4231206A1Method for multi object tracking with quantum adiabatic computing
Publication Date: 2023.08.23 TOYOTA JIDOSHA KK

AI summary

A computer-implemented method for grouping detected objects (O1-OD) such that a group (Grt) comprises detected objects (Od) corresponding to a given type of object, a said detected object (Od) being on a frame (Fi) belonging to a sequence of frames (F1-FN), said method consisting in determining values (Oval) of assignments (As11-AsDG) of said detected objects (O1-OD) to said groups (Gr1-GrG), said method comprising the following steps: - obtaining (EOC), for each pair of detected objects (Odi, Odj), a similarity score (Simdidj); - determining (EDQ), based on said similarity scores (Simdidj), an objective function (QCF) quadratic with respect to said assignments; - evaluating (EOp) values of said assignments, said evaluating step comprising the following substeps: (i) obtaining (EOp10), using a quantum annealing solver (QAS), an estimated solution (Sol) of an unconstrained binary quadratic optimization problem (QUBO) corresponding to said objective function (QCF); (ii) post-processing (EOp20) said estimated solution (Sol) to obtain said values (Oval) of assignments (AS11-AsDG).