Unmanned Systems Technology 021 | Robot Aviation FX450 l Imaging Sensors focus l UAVs Insight l Liquid-Piston X-Mini l Riptide l Eurosatory 2018 show report l Zipline l Electric Motors focus l ASTS show report

35 the image quality isn’t as good, but the parallel output gives a higher bandwidth and the output has less electrical noise. Machine vision algorithms used for unmanned systems are less sensitive to image quality but need the higher bandwidth and lower noise. The latest CMOS technology has features as small as 90 or even 65 nm; it is well-suited to image sensors and is mature compared to the leading-edge 10 and 7 nm technologies. That means the CMOS sensors can be made more cheaply than CCD and with lower power. This is critically important with multiple sensors being used in a vehicle. However, the choice of sensor also depends on the application. CCD sensors can be fabricated with a thicker sensing layer, called the epitaxy or epi layer, for IR sensing while still having a reasonable resolution. This epi layer can be more than 100 microns thick to collect more charge from the photons that hit the sensor, compared to the 5-10 microns in a CMOS sensor. This gives a CCD sensor greater sensitivity. Another challenge is that the design of the sensor systems is becoming more closely coupled to the image processing algorithms, and these algorithms vary from one car maker to another. For example, the move to a global shutter design is being driven by the increasing use of machine vision and machine learning algorithms to automatically identify elements in a picture. The sensors also have to be able to cope with emerging roadside technology, such as LED signs and the increasing use of LED headlights that operate across a wide range of frequencies. Both of these can cause problems for some image sensing technologies. In the air there are two different uses for the sensors. UAVs are increasingly using them for navigation and collision avoidance, while the demands for high- definition video and still images from cameras is also pushing performance. Payloads have always included thermal imaging cameras and hyperspectral cameras, and these continue to improve in performance to deliver the data that users want. Autonomous vehicles One of the biggest challenges for the sensors in cars these days is their low- light performance. Making pixels in the sensor larger to collect more of the available photons and give more sensitivity in low light makes the sensors larger and more costly, as well as increasing the electronic noise in the system. It also takes longer for the pixel to trigger, making the sensor slower. They also have a lower dynamic range, as they are more easily overwhelmed in bright conditions. So there is a delicate balance between the size of the pixel, the read noise and speed of the sensor, and the high dynamic range (HDR). Automotive systems require an HDR of 100-120 dB for situations with rapidly changing light conditions, such as emerging from a dark tunnel into bright light. LED signs and headlights will always be running at an unknown and unspecified frequency, and an image sensor optimised for HDR will tend to flicker as a result. A new generation of sensors has been designed with LED flicker mitigation, however, which is critical for the machine vision algorithms, for example detecting whether a brake light has come on or reading speed limit signs. These requirements have resulted in specially designed pixels with an overflow capability that gives the pixel the ability to store more Imaging | Focus LED headlights will be running at an unknown frequency, and an image sensor optimised for HDR will tend to flicker as a result Unmanned Systems Technology | August/September 2018 Pixel arrays are mounted on top of the analogue and digital processing chip to reduce size and increase bandwidth (Courtesy of On Semiconductor)