[Paper Review] MILD: Multi-Index Hashing Based Loop Closure Detection

Lei Han, Lu Fang|arXiv (Cornell University)|Feb 28, 2017

Robotics and Sensor-Based Localization18 references3 citations

TL;DR

MILD proposes a real-time loop closure detection method using Multi-Index Hashing (MIH) to enable high-recall image similarity matching via direct feature matching, avoiding the centroid ambiguity of traditional binary bag-of-words approaches. It achieves superior efficiency and accuracy by theoretically balancing approximation accuracy and computational cost through probabilistic analysis of MIH.

ABSTRACT

Loop Closure Detection (LCD) has been proved to be extremely useful in global consistent visual Simultaneously Localization and Mapping (SLAM) and appearance-based robot relocalization. Methods exploiting binary features in bag of words representation have recently gained a lot of popularity for their efficiency, but suffer from low recall due to the inherent drawback that high dimensional binary feature descriptors lack well-defined centroids. In this paper, we propose a realtime LCD approach called MILD (Multi-Index Hashing for Loop closure Detection), in which image similarity is measured by feature matching directly to achieve high recall without introducing extra computational complexity with the aid of Multi-Index Hashing (MIH). A theoretical analysis of the approximate image similarity measurement using MIH is presented, which reveals the trade-off between efficiency and accuracy from a probabilistic perspective. Extensive comparisons with state-of-the-art LCD methods demonstrate the superiority of MILD in both efficiency and accuracy.

Motivation & Objective

• To address the low recall issue in binary feature-based loop closure detection caused by poorly defined centroids in high-dimensional binary descriptors.
• To enable real-time performance in loop closure detection without sacrificing recall.
• To develop a method that directly matches features using Multi-Index Hashing to avoid reliance on centroid-based representations.
• To theoretically analyze the trade-off between accuracy and efficiency in approximate image similarity measurement using MIH.
• To demonstrate superior performance in both efficiency and accuracy compared to state-of-the-art LCD methods.

Proposed method

• MILD employs Multi-Index Hashing (MIH) to index binary feature descriptors, enabling efficient and approximate similarity search.
• It directly matches features using MIH instead of relying on centroid-based representations, thereby improving recall.
• The method uses a probabilistic framework to analyze the trade-off between approximation accuracy and computational efficiency in MIH-based similarity measurement.
• MIH constructs multiple hash tables, each using a different random projection, to improve matching robustness and reduce false negatives.
• Image similarity is computed via intersection of hash codes across multiple indexes, enhancing retrieval precision.
• The system is designed for real-time operation by minimizing computational overhead while maintaining high recall.

Experimental results

Research questions

• RQ1 How can loop closure detection achieve high recall without increasing computational complexity?
• RQ2 What is the theoretical trade-off between accuracy and efficiency in approximate image similarity measurement using MIH?
• RQ3 Can direct feature matching via MIH outperform centroid-based bag-of-words methods in recall and speed?
• RQ4 How does MIH-based similarity measurement compare to state-of-the-art LCD methods in real-world SLAM scenarios?
• RQ5 What is the impact of multi-indexing on the robustness and recall of loop closure detection?

Key findings

• MILD achieves higher recall than centroid-based binary feature methods by avoiding the ambiguity of high-dimensional binary descriptor centroids.
• The method maintains real-time performance due to the low computational overhead of MIH-based feature matching.
• Theoretical analysis confirms a quantifiable trade-off between approximation accuracy and efficiency in MIH, guiding optimal parameter selection.
• Extensive comparisons show MILD outperforms state-of-the-art LCD methods in both accuracy and speed.
• Direct feature matching via MIH significantly improves retrieval performance compared to traditional bag-of-words approaches.
• The use of multiple hash indexes enhances matching robustness and reduces false negatives in loop closure detection.