HDR mode of CMOS sensors
1. Introduction
Industrial cameras used outdoors face particular challenges: Changing light conditions through day and night and if they are used in traffic surveillance, backlighting can be a common problem. High dynamic range of the sensor is the way to go and is often wrongly associated with expensive. The AR0331 CMOS sensor from Aptina/ON Semiconductor with HDR mode proves that this is not the case. The following document shows what it can do and how it is operated.
2. HDR
The abbreviation HDR stands for "High Dynamic Range" and describes a video or single image transmission chain that can reproduce the greatest possible differences between light and dark in an image. For this reason, HDR is also a current buzzword for new flat-screen televisions. In other words, since black can no longer be made blacker, white must be made brighter. Figuratively speaking, the viewer should have to squint when the protagonist's flashlight is shone through the television into the living room. This is by no means the case today. The following Wikipedia table provides an overview of transferable dynamic range values:
Dynamic range of common devices |
||
Device |
Stops |
Contrast |
LCD |
9,5 |
700:1 (250:1-1750:1) |
Negative film (Kodak VISION3) |
13 |
8000:1 |
Human eye |
10-14 |
1000:1-15000:1 |
High-end DSLR camera (Nikon D810) |
14,8 |
28500:1 |
If one assumes the intuitively understandable contrast ratio of light to dark, for example 8000:1, the logarithmic representation results from 20*log(8000) = 78 dB. As each bit now brings ~6 dB (factor 2) of dynamic range, 78 dB/6 = 13 bits of dynamic range, which is equivalent to the number of f-stops, each of which also contributes to a doubling of the signal.
It can be seen that the human eye is still the benchmark and is just about matched by a professional SLR camera and, for some time now, by chemical negative film. The image display on a flat screen usually does not achieve these contrasts at present.
3. Dynamics of machine vision cameras
The EMVA1288 standard makes it possible to measure and compare the dynamic range. A Sony IMX sensor from the Pregius series achieves over 70 dB, which corresponds to ~12 bits. Furthermore, high-quality CCD sensors from Sony achieve around 60 dB (10 bit) and inexpensive rolling shutter sensors achieve over 60 dB. More is not normally possible with machine vision cameras at present.
4. How is HDR feasible for live images?
Driven by the innovation in sensors for cell phone cameras, which produce extremely good images despite small pixels and rolling shutters, the AR0331 sensor from Aptina/ON Semiconductor offers the following interesting option: While the sensor already achieves a very good dynamic range of ~70 dB with its 12-bit analog-to-digital converter (ADC), this rolling shutter sensor can also operate two interleaved reset → exposure pointers. This functionality can be used to generate a 16-bit linear signal in which the two exposure times are in certain ratios. So if the shorter exposure time is 1/16 of the long one, the upper 4 bits can be added to the lower 12 bits of the long integration time to create a 16 bit value.
AR0331 has 12 bit ADC (~72 dB) 100 dB requires ~16 bits
A 16-bit linear image, which was generated from an exposure with 4 ms and a second exposure with 250 µs, now looks as follows as an example:
From the top left to the bottom right, each shift is shifted down by one bit, i.e. the exposure values are doubled with each shift. You can see that the lamp is least overdriven in the upper bits and that the remaining image content can only be recognized in the lower bits.
5. So what are the challenges?
The most important thing is certainly to make this image with its high dynamic range look natural on a display that cannot show the high dynamic range according to the initial table. This process is called Adaptive Local Tone Mapping (ALTM). This is achieved, for example, using a special adaptive companding characteristic on the sensor:
This creates the following image, which combines the lamp and the darker image content in one image.
Additional functions such as motion compensation and the avoidance of color clipping effects ensure that even moving images can display a higher dynamic range, preferably without disturbing artifacts. Mind you, everything is automatic and at the full frame rate of 30 frames per second at Full HD resolution and in perfect color quality. The AR0331 from Aptina/ON Semiconductor is used in the USB3 Vision camera with the designation BVS CA-SF1-0031ZC. There are also further "fine-tunings", in that the sensor registers can be accessed directly via the driver and options can be switched on or off.