Metrological comparison between a generalised N-dimensional classical and quantum point cloud

Massive Analytic Ltd

National Physical Laboratory
Round 3

The large-scale multimodal sensor fusion of loT data can be transformed into a classical point cloud (CPC).

For example, the transformation may be the fusion of three imaging modalities of different natures such as LiDAR, a set of RGB images, and a set of thermal images.

To combine imagery from different data sources MAL uses a variety of SLAM (Simultaneous localization and mapping) techniques for sensor fusion including scale-invariant feature transforms (SIFT), occlusion detection, manifold fusion, Extended Kalman filters, and visual odometry.

However, it is not easy to process a point cloud, because in order to record the surface clearly, a point cloud usually has millions or even hundreds of millions of points.

As a result, classical computers often crash when operating a point cloud of multimodal sensor data.

Quantum computers are expected to solve the multimodal sensor point cloud processing problem more efficiently.

While for a long time quantum computers were only a vision for the distant future, they now exist with the number of available qubits steadily increasing (current quantum computers have about 50 qubits).

In order to operate point clouds in quantum computers, there are two problems to be solved; quantum point cloud (QPC) representation and quantum point cloud processing (QPP).

Fused multimodal sensor data can be represented as an N-dimensional quantum point cloud.

A numerical solution suitable for quantum point cloud processing can be derived.

This project is a metrological evaluation involving the comparison of various measures between a CPC and a QPC, including how long does it take to prepare and process the quantum point cloud and the accuracy and uncertainty of the results.

A CPC of multi-modal sensor data is the starting point, analysed with MAL’s artificial precognition ML algorithms to measure data quality such as point density, point numbers, shadow removal, noise, precision, uncertainty and errors.

It will then be ‘improved’ and made ready for quantum processing by compressing it first to reduce the number of quantum gates required.

NPL will generalize published QPC 3D quantum algorithms to N-dimensional input data and evaluate scalability, accuracy and uncertainty.

Numerical simulations will be performed on a quantum simulator.

The project will explore how an N-dimensional QPC addresses the problem of uncertainty in multi-modal sensor data, such that models built using MAL’s artificial precognition machine learning algorithms can be derived with outcomes of greater certainty – drawing a direct comparison with the CPC.

Analysis for Innovators