Dropcam Echo : Home Security Meets the Cloud

It has been quite some time since AnandTech reviewed cameras or camcorders for their image and video quality. Even amongst professional gadget review sites, there is an absence of a standardized benchmarking methodology for this purpose. Starting with this review, we have decided to use the Handsome Case for evaluating image and video capture gadgets. This kit can be used to determine and analyze many image and video quality characteristics. However, this section will only deal with factors relevant to a security camera in a home setting.
 

It is very important for any security camera to have a good performance in low lighting conditions. We performed a quick image quality test of the Dropcam at various light intensity levels. The light temperature was held constant at 5500 K and three light levels were evaluated: 1000 lux, 260 lux, and 15 lux. The evaluation was not extensive and did not included moving objects or resolution measurements. In general, the camera held up well under low light conditions in this static scenario.

One interesting thing is the choice of holding color saturation levels somewhat constant under lower light conditions. Chroma noise becomes prominent at 15 lux, especially noticeable in dark low frequency areas. It is our understanding that most industrial security applications make the design decision to reduce chroma as light intensity falls or simply to move to black and white mode at a given light intensity. The typical security professional would prefer less noise or more color to maintain resolution and detail. Since this is a consumer targeted unit we believe they made the right decision as most consumers likely prefer to see color and will accept the tradeoff of chroma noise. That being said, it might be nice to have a "black and white mode" or some configurability in this area for the discerning low light viewer.

Again, no detailed analysis was performed on the resultant bitstream but the smaller block sizes present in the lower lux scenes seems to indicate that more bits are being used to encode these scenes as the encoder will need to work harder to encode the increased noise patterns.
 

In Figure 6A, note the low frequency dark area of the color meter. As light intensity is decreased this area contains more noise - especially in the chroma plane. The encoder interprets the noise as detail that needs to be encoded. We don't know the internal workings of the image pipeline, but a 3D noise filter that identifies spatial noise patterns over time would certainly help here. Such filters are present in most of the present day IP camera chipsets. So, future models are likely to have improvements in this area.
 

In Figure 6B, note the red feather in the scene captured at high lux with a DSLR compared to the screen shots of the unit encoding at 1000 Lux and 15 Lux. The unit loses most of the fine detail at 1000 lux and then works much harder in the lower lux environment likely because a moving noise pattern is present and this noise is attempting to be encoded.
 

The human eye is trained to look carefully at faces. In Figure 6C, some degradation of the facial resolution is noticeable between 1000 lux and 15 lux, but skin tone is maintained nicely. Note the difference in chroma noise in the low frequency dark area behind each doll.

To summarize the DSP qualities, it can be said that the unit holds up well under differing low light conditions. Chroma levels are maintained with the resultant artifacts noticeable. More configuration options would be nice for low light scenes but the target audience is probably going to be happy with the tradeoffs in the shipped unit.