Lidar technology breakthrough wins top paper prize

 Figure shows 3D imaging performance of a Lidar camera in a setting with strong background illumination. Top left: an image processed “pixelwise” – or pixel by pixel – results in a “noisy” image, with less definition and accuracy. Image quality can be substantially improved with spatial regularization algorithms such as RT3D (top right). However, in both cases the data transfer involved is prohibitive. Bottom row: the pixelwise sketched Lidar reconstruction (left), and the sketched RT3D images (right). In b
Figure shows 3D imaging performance of a Lidar camera in a setting with strong background illumination. Top left: an image processed “pixelwise” – or pixel by pixel – results in a “noisy” image, with less definition and accuracy. Image quality can be substantially improved with spatial regularization algorithms such as RT3D (top right). However, in both cases the data transfer involved is prohibitive. Bottom row: the pixelwise sketched Lidar reconstruction (left), and the sketched RT3D images (right). In both cases the results are almost identical to those achieved with full data processing, but require less than 1% data transfer.

A paper on “How to process billions of photons a second” for ultrafast and accurate 3D imaging by Dr Julián Tachella, Dr Michael Sheehan and Professor Mike Davies has won Best Student Paper award at ICASSP 2022, the top international signal processing conference.

Self-driving cars rely on detailed and precise real-time 3D imaging, in order to “see” their surroundings, steer safely and avoid collision.

These images are produced by Lidar (laser sensing) devices, which need to track billions of photons a second to achieve accurate high-resolution images. This generates an enormous amount of data for processing, which can limit the accuracy and resolution images produced, restricting the deployment of real-time systems.

In the paper, the researchers showed how a new data acquisition method can be employed to produce highly accurate real time 3D images while simultaneously compressing the data by more than 99%, without sacrificing image quality.

This significantly reduces the memory and computational power requirements in Lidar devices, removing key barriers to fully self-driving cars.

The Edinburgh team is currently collaborating with Ibeo Automotive Systems, a worldwide leader in the field of Lidar sensors, to develop commercial applications of this technology.

The research team

Dr Sheehan recently completed his PhD at the University of Edinburgh under Professor Davies’ supervision, while Dr Tachella was formerly a Postdoctoral Research Associate in Professor Davies’ research group in the School’s Institute for Digital Communications.

Find out more