'Boaring' in on Vehicle-To-Animal Road Safety
Researchers are using micro-Doppler radar, neural nets and machine learning to protect drivers from the wild critters that enter the road.
A wild boar can ruin a driver’s day — and a lot more. Hitting one of these fast-running (up to 25 mph/40 km/h) and heavy (up to 400 lb/181 kg in the southern U.S.) porkers when they dart into a vehicle’s path is a serious threat. Indeed, reducing vehicle-to-animal crashes is a global safety concern, one that is the focus of a novel pioneering research study in Europe.
The project, underway at two German universities, aims to classify and differentiate the potential behavior of moving hazards. The researchers are using micro-Doppler radar complemented by cameras and infra-red (IR) sensors, together with neural networks allied to algorithms developed in a machine-learning system. The combination would warn drivers and other road users of dangers ahead.
The accretion of real-time intelligence enables the system to foresee how a hazardous situation could develop. Not limited to wild boar, the project’s list of potentially menacing wildlife includes deer and foxes – to be accurately and reliably differentiated from pedestrians, cyclists, cars and motorcyclists.
“As far as I know, radar sensors have not been used to detect wildlife,” observed Prof. Dr. Hubert Mantz, one of the project leaders at Ulm University of Applied Sciences. He noted that both optical and thermal imaging cameras have had success in this area. Now the demonstrator project, called SALUS, is working towards a broader application of technology to provide enhanced information. Ulm University is working with Heilbronn University of Applied Sciences and with industrial partners including Frankfurt-based Silicon Radar. Data from the micro-Doppler radar is gathered by a Spectrum Instrumentation digitizer.
SALUS is an acronym whose English translation is “safe and sustainable mobility for urban and rural regions through intelligent road structures.” It’s also the name of the Roman goddess of safety. “Our approach is to record different patterns of moving animals and to train a learning system in such a way that other patterns of moving animals can be detected as ‘similar’ and thus dangerous for road traffic,” said Mantz. “This is ‘new’ and has not been done before.
“We have a setup that basically can collect two data sets simultaneously,” explained Mantz. “Firstly, a time-Doppler dataset; this shows the velocity distribution during the measurement time. This can be seen repeating patterns in time — for example, a walking animal or human has a periodicity that can be seen in time-Doppler plots. Secondly, we can record a range-Doppler plot, which shows the velocities in a certain distance from the sensor. This can be recorded continuously so that we also have a time-resolution.”
The Spectrum Instrumentation digitizer used to collect data from the micro-Doppler radar is a PCLe digitizer M2p.5926-x4 with 16-bit, four differential channels and 10 MHz bandwidth. It gathers the information to teach the system as it enables the team to simultaneously process all the data needed in real time. “We can focus on the project rather than programming,” noted Mantz.
The basis of the project’s wildlife work started via recording data at a zoo, a deer park and in fields with a flock of sheep and sheepdogs. That helped the team clearly identify characteristic walking patterns based on time-Doppler and range and Doppler spectrograms of the animals observed. A further goal is to distinguish between animals presenting a danger to road users, and other extraneous movements such as windblown trees.
The SALUS project also is examining the use of what it terms roadside “small installations” – solar-powered posts - that could be widely deployed to detect hazards and communicate alerts to vehicles. In many areas, wildlife commonly have specific road crossing points, so such “inexpensive” posts could be placed in specific areas. The German research teams also are examining the possibility of using the system to measure pollution levels.
Mantz explained that the use of neural networks to develop machine learning to achieve differentiation between pedestrians, cars and wildlife “takes it far beyond pure motion detection.” He said: “We are now at the critical part of the project – the classification of detected objects, which has never been done before.” The boars may be pleased to know they will be safer, too.