Multiple Object Tracking Initialization via Consistency Error Index

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing multiple object tracking methods, such as SORT and DEEPSORT, face challenges in initializing tracking identifiers quickly, especially when image acquisition frequency is low relative to the dynamics of moving objects, and are computationally expensive, leading to delayed or inefficient tracking.

Innovation Solution

A method that involves receiving a sequence of images, detecting objects using a convolutional neural network, determining bounding boxes, associating objects with unique tracking identifiers based on a minimum overlap threshold, calculating a consistency error index for candidate sequential boxes, and assigning identifiers to those with the lowest error index, allowing rapid initialization and robust tracking even with non-consecutive images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If geometric initialization methods (SORT) are used to track objects, then the tracking algorithm runs in real-time, but tracking identifiers are not initialized quickly enough when image acquisition frequency is low compared to object dynamics

Engineering Contradiction:
Improveobject dynamics speedVSAvoidinitialization time
Core Design Contradiction:
SpeedVSLoss of time

Solution Approach 1:

The patent applies preliminary action by using a buffer to store multiple consecutive images (not just the current image) before processing. This allows the system to have multiple frames available in advance for initialization attempts, enabling faster identification and tracking of newly entered objects without waiting for the standard frame-by-frame processing sequence.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If deep learning methods (DEEPSORT) are used for hybrid geometric initialization, then tracking identifier initialization is improved, but the computational cost increases significantly making tracking slower

Engineering Contradiction:
Improvetracking identifier initialization accuracyVSAvoidtracking speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent changes the parameter of image sequence length from the standard single-frame or two-frame processing to processing a buffer of multiple consecutive images (N frames). This parameter change allows the system to accumulate sufficient object movement information across multiple frames to reliably initialize tracking identifiers using geometric methods, achieving DEEPSORT-like accuracy without the computational overhead of deep learning.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4439478A1Method for tracking multiple objects
Publication Date: 2024.10.02 IDEMIA ROAD SAFETY FRANCE
  • EP4439478A1 patent drawingFigure 1
  • EP4439478A1 patent drawingFigure 2~3
  • EP4439478A1 patent drawingFigure 4

AI summary

Method for tracking multiple objects in temporal image acquisitions, comprising for the initialization of the tracers: - a step of determining the bounding frame of an object detected without an associated identifier among the images; - a step of determining a set of candidate sequential multiplets of said bounding frames (d1, d2, d3) of an object detected without an identifier, each multiplet comprising as its first element a frame (d1) without an identifier of an image h, as its second element a frame (d2) without an identifier of an image i, with i>h and as its third element a frame (d3) without an identifier of an image j with j>i; - a step of calculating a coherence error index for each multiplet; the coherence error index being calculated as a function of k-1 vectors (v1, v2) each linking the reference point of one element to the reference point of the next element.