Possibilities for system design with ultra-low power cameras
An energy efficient system is a key design requirement for various applications especially for Internet of Things (IoT) applications. This involves setting up a camera system where a power supply isn’t available. An application area is the adoption of Body Worn Cameras (BWC`s) by law enforcement officials. The “always-on”- mode puts heavy constraints on battery usage.
In this story, we will just look at requirements and possibilities for an ultra-low power system design.
The best sensor for your application: CCD vs. CMOS
CCD Sensors are good to use for large sizes with small pixels. Furthermore, through a passive charge transfer there is less Fixed Pattern Noise (FPN) and no time smear. If you have big pixel arrays, you have high power consumption because of the readout architecture.
The CMOS technology has the advantage that all readout elements can be integrated monolithically on the same chip. If they have an active pixel sensor, they can intensify the signal and buffer them by their own. The CMOS sensors can convert the photocurrent into an electrical signal before being transferred to a common leader. This Random Access Readout offers the potential for high-speed readout at low power consumption.
To reduce the data volume of your system data compression is indispensable nowadays. A basic distinction is made between lossless and lossy video compression. Lossless video compression is more efficient but the decoded signal no longer looks like the original signal. The most commonly used video encoders are MPEG-4 and H. 264/AVC, where H.264/AVC defines a standard that specifies the type of representation and decoding method.
H.264/AVC is good in compression but has a higher power consumption due to it’s high complexity. The efficiency is still better than MPEG-4.
PWM and voltage scaling
Voltage scaling reduces the consumption down to <0.5V. For CMOS sensors, you can use lower energies for pixel arrays. Due to less voltage fluctuation and therefore a lower signal level, a bad signal-to-noise ratio can easily occur.
Good sensors with a high resolution and a high light sensitivity have a high power consumption. The sensor has to be adapted to the environmental conditions so there are different modes: when there is no movement they are in a “sleep-mode” and use less energy, if there is a movement in the field of view, they use the full bandwidth and have a high resolution.
The energy harvesting captures stray energy from the environment into electrical energy. Solar energy is widely used and saves up to 100 mW/cm2 so it is the most efficient way. However, it can only be used if there is enough daylight. Human motion and vibration are also potential energy sources but not so much energy can be harvested in comparison to solar. So normally it isn´t worth.
Dynamic Vision Sensor
Conventional image sensors are frame based and transmit at a constant frame rate. This generates high voltage consumption and may cause a delay in data processing or even a loss of information if the movement is faster than frame time.
A Dynamic Vision sensor, on the other hand, has a frame time of microseconds and shows only the changes in a scene. The pixels react to changes in brightness and were read out individually so that the processor is less affected. This results in a lower power consumption of up to 100 times lower than with conventional sensors.
Review and available sensors
Today there are no sensors, which use PWM and voltage scaling efficiently for our application. There are some sensors which use them, but they are often specialized for an application.Voltage scaling implies strong noise and PWM can compensate that a little bit but the signal to noise ratio is still bad.
Event control is even more efficient. The sensor from Himax for example is event controlled. The datasheet is linked below. This sensor is especially good if it`s used for an application where only sometimes something happens and otherwise the sensor is in sleep-mode. This sensor is appropriate for end-users, medical and industrial applications. The Himax sensor uses 4mW in sleep-mode. The power consumption in continuous operation would be higher.
A Dynamic Vision Sensor implies that we can use only monochrome information. This is in some application a big disadvantage so that it can´t be used. If not that is a good choice because of the high savings in power consumption. The sensor from Omni Vision for example is also a DVS sensor but is also available in color. This sensor offers enhanced functions like gesture control and motion detection and is appropriate for smartphones, notebooks and tablets. This sensor need 87mW and is the only sensor, which offers Full High Definition (FHD) quality, what means 1920 x 1080 pixels.
Omni Vision: http://www.ovt.com/sensors/OV2740
Energy harvesting is useful if there are possibilities to get stray energy from the environment. For our application, we need a high resolution this causes a high power consumption. For this case there is no available sensor, which uses photovoltaic, especially because it isn´t useful for body worn cameras which should also operate at night.
Sony offers a sensor, which could be the smallest 1MP sensor. This sensor offers a really good image quality at a very small size. This Sensor is appropriate for IoT, wearables and drones.
Project: body-worn Cameras with an ultra-low power sensor and an operating time about 10 hours
For the project we choose the Sony sensor because of its availability. Basically the sensor from Omni Vision is better in resolution and the power consumption per frame per pixel (called FoM). But this sensor wasn’t readilyavailable. Furthermore, we have an evaluation board with a controller from Ricoh and a Raspberry Pi 3.
The sensor consumes about 55mW at a resolution of 1280x720 and 60 fps. An SD-card with 32 MB should last long enough if we film a video about 10 hours with H.264 compression and audio. The battery is a smartphone battery from the Lenovo P2, this battery has a capacity of 5000 mAh.
The picture shows the construction:
We used the software Vision Processing Framework to control the functions of the sensor. The sensor is used as webcam over USB 2.0. The recording a compression is done by the FFmpeg software.
Overall, it is possible to design a very small system, which operates with ultra-low power. However, you have to look on the real power consumption if the camera operates in the environment because this a function of the complexity of the picture operating temperature and frame rate.