[Paper Review] Multi-objects association in perception of dynamical situation

Dominique Gruyer, V. Berge-Cherfaoui|arXiv (Cornell University)|Jan 23, 2013

Multi-Criteria Decision Making7 references20 citations

TL;DR

This paper proposes a belief-theory and fuzzy-logic-based multi-object association algorithm for dynamic environment perception in intelligent vehicles, enabling robust tracking under sensor uncertainty and ambiguity. By fusing numerical and symbolic data and resolving conflicts via an assignment algorithm, it achieves reliable object association even during object appearance/disappearance or confusing perceptions, significantly improving environment map accuracy and tracking stability.

ABSTRACT

In current perception systems applied to the rebuilding of the environment for intelligent vehicles, the part reserved to object association for the tracking is increasingly significant. This allows firstly to follow the objects temporal evolution and secondly to increase the reliability of environment perception. We propose in this communication the development of a multi-objects association algorithm with ambiguity removal entering into the design of such a dynamic perception system for intelligent vehicles. This algorithm uses the belief theory and data modelling with fuzzy mathematics in order to be able to handle inaccurate as well as uncertain information due to imperfect sensors. These theories also allow the fusion of numerical as well as symbolic data. We develop in this article the problem of matching between known and perceived objects. This makes it possible to update a dynamic environment map for a vehicle. The belief theory will enable us to quantify the belief in the association of each perceived object with each known object. Conflicts can appear in the case of object appearance or disappearance, or in the case of a confused situation or bad perception. These conflicts are removed or solved using an assignment algorithm, giving a solution called the " best " and so ensuring the tracking of some objects present in our environment.

Motivation & Objective

• To address the challenge of associating perceived objects with known objects in dynamic environments for intelligent vehicles.
• To enhance environment perception reliability by managing uncertainty and ambiguity from imperfect sensors.
• To develop a robust association mechanism that maintains tracking during object appearance, disappearance, or confusion.
• To integrate numerical and symbolic data through belief theory and fuzzy mathematics for comprehensive perception.
• To resolve conflicts in object association using an assignment algorithm to ensure a consistent and optimal tracking solution.

Proposed method

• The algorithm employs Dempster-Shafer evidence theory to quantify belief in object associations, handling uncertainty and imprecision.
• Fuzzy mathematics is used to model sensor data and object features, enabling representation of inaccurate or vague information.
• A belief-based matching process computes the degree of belief that each perceived object corresponds to each known object.
• Conflicts arising from ambiguous perceptions or sensor noise are resolved using an assignment algorithm to select the optimal association set.
• The system dynamically updates the environment map by integrating confirmed object associations over time.
• The fusion of numerical (e.g., position, velocity) and symbolic (e.g., object type) data is natively supported through the belief-structure framework.

Experimental results

Research questions

• RQ1 How can object associations be reliably maintained in dynamic environments with uncertain and imprecise sensor data?
• RQ2 What mechanisms can resolve conflicts in object association when multiple perceived objects could match a single known object?
• RQ3 How can both numerical and symbolic data be effectively fused in a perception system for intelligent vehicles?
• RQ4 What role does belief theory play in quantifying uncertainty during object tracking?
• RQ5 How can the system maintain consistent tracking during object appearance, disappearance, or confusion?

Key findings

• The proposed algorithm successfully resolves conflicts in object association using an assignment algorithm, ensuring a consistent and optimal tracking solution.
• The integration of belief theory and fuzzy mathematics enables effective handling of uncertain, imprecise, and ambiguous sensor data.
• The system maintains reliable object tracking even during object appearance or disappearance, which are common challenges in dynamic environments.
• The fusion of numerical and symbolic data is achieved within a unified framework, improving the robustness of environment perception.
• The method demonstrates improved reliability in environment map updating by quantifying belief in associations rather than relying on deterministic matching.
• The approach is validated in the context of intelligent vehicle perception, showing applicability to real-world dynamic scenarios with sensor limitations.