Optimising Radar Training with Machine-Assisted Machine Learning
The Challenge
A common limitation when developing new AI models is the availability of labelled ground truth data. This can require significant resources to collect and label before the machine learning process can even begin. This is a particular problem when developing a target classification algorithm for a new radar.
What if a pre-existing model and a camera could be used to label the radar training data automatically?
The Approach​
An open-source object detection algorithm can be used with a standard security camera to detect and classify objects in the scene. If this camera is co-located with the radar, these labels can also be applied to the simultaneously collected radar data. Leaving both sensors viewing a road unattended for an extended period time, produces a radar data set with labels for foot and road traffic of various types. This data set can then be used for AI model training.
A neural network classifier was then trained from scratch using the raw radar data and associated ground truth labels. As is common with radars, the data consisted of the measured intensity as a function of range, azimuth angle and Doppler velocity. This custom built Plextek radar was designed to have good resolution across all of these dimensions:
- 40 cm range resolution
- 1.8° azimuth resolution
- 0.5 km/h Doppler resolution
Please accept cookies in order to enable video functionality.
The Outcome
The resulting radar classifier successfully discriminated different types of vehicles and pedestrians to allow automatic traffic monitoring. While the colour camera could classify traffic during the day when lighting is sufficient, the radar classifier can operate both day and night as it is unaffected by lighting conditions. Additional training data can also be generated without excessive personnel requirements.
Revolutionise your radar training
At Plextek, we understand the challenges faced in developing AI models, for example the labour of collecting and labelling ground truth data for radar training. Our innovative approach combines an open-source object detection algorithm with a standard security camera to automatically label radar data, saving valuable time and resources.
By partnering with us, you could benefit from improved accuracy and efficiency in AI model development, while significantly reducing the resource needed for generating additional training data.
Our team of exceptional engineers can help revolutionise your AI model training process with cutting-edge technology – enabling you to set new standards in radar data processing and classification.
Enhancing network efficiency with ML & RL
Enhancing Mobile Ad-Hoc Network Routing with GNN-Driven DRL Agent: Improving Packet Delivery and Adaptability in Dynamic Environments
Read More
Semantic Segmentation in Visual Scene Matching
Learn how we boost accuracy in Visual Scene Matching with Semantic Segmentation, aiding navigation and tracking in dynamic environments.
Read More
Machine Learning for Rapid Propagation Assessment
Developing a groundbreaking ML model for swift and efficient coverage prediction in complex urban environments, enabling rapid optimisation of transmitter locations on standard computing hardware.
Read More
Future Sensing: Improving Mobile Ad-hoc Networks
Leading a transformative four-year research initiative to improve mobile ad-hoc networks through advanced directional antenna systems and cross-layer processing, significantly enhancing military communication capabilities.
Read More
Armour Integrity Monitoring System (AIMS)
Innovating a new Armour Integrity Monitoring System (AIMS) for the UK MoD, delivering a low SWaP-C solution that dramatically streamlines logistics and enhances protection through in-field armour integrity checks.
Read More
mmWave Radar for Foreign Object Debris Detection
Collaborating with WaveTech to develop an advanced mmWave radar system, enabling the rapid and automated detection of foreign object debris on runways, enhancing safety and operational efficiency at a South Korean airport.
Read More
Surveillance Radar for Comprehensive Threat Detection
Advancing a perimeter surveillance solution with long-range detection and low false-alarm rates, using state-of-the-art Passive Electronically Scanned Array technology for robust and maintenance-free operation in a range of demanding environments.
Read More
Immersive Technology for Complex Systems Training
Using the latest immersive technologies to deliver realistic and engaging training solutions for complex systems, ensuring specialist procedures are effectively imparted and measured within any environment.
Read More
Distributed Real Time Spectrum Monitoring
Developing an innovative distributed spectrum monitoring system, using low-cost software-defined radio platforms, to provide superior interference detection and larger coverage for high-value sites.
Read More
Intelligent Mobility
Advancing intelligent mobility by integrating cutting-edge electronic-scanning radar technology to ensure the safe and efficient operation of autonomous vehicles in complex real-world environments.
Read More
Telehealth Innovation – Connecting Patients to their Carers
Revolutionising telehealth with the MonitorMe phone, seamlessly connecting patients to healthcare providers for vital signs monitoring.
Read More
Drone Inspection of Solar Farms
Enhancing solar farm efficiency through advanced drone inspection and automated analysis of diagnostic data.
Read More
Related Technical Papers
View All
Non-Invasive Auditory Sensing with Affordable Headphones
This paper presents a sensor for measuring auditory brainstem responses to help diagnose hearing problems away from specialist clinical settings using non-invasive electrodes and commercially available headphones. The challenge of reliably measuring low level electronic signals in the presence of significant noise is addressed via a precision analog processing circuit which includes a novel impedance measurement approach to verify good electrode contact. Results are presented showing that the new sensor was able to reliably sense auditory brainstem responses using noninvasive electrodes, even at lower stimuli levels.
GPU Computing
Power limits restrict CPU speeds, but GPUs offer a solution for faster computing. Initially designed for graphics, GPUs now handle general computing, thanks to advancements by NVIDIA, AMD, and Intel. With hundreds of cores, GPUs significantly outperform CPUs in parallel processing tasks. Modern supercomputers, like Titan, utilize thousands of GPU cores for immense speed. NVIDIA’s CUDA platform simplifies GPU programming, making it accessible for parallel tasks. While GPUs excel in parallelizable problems, they can be limited by data transfer rates and algorithm design. NVIDIA’s Tesla GPUs provide high performance in both single and double precision calculations. Additionally, embedded GPUs like the NVIDIA Jetson TX2 deliver powerful, low-power computing for specialized applications. Overall, GPUs offer superior speed and efficiency for parallel tasks compared to CPUs.
