Simplicam Facial Recognition Tested
Facial recognition, available for $150?
That's the offer from a startup, Simplicam, [link no longer available] who has not only cloned Dropcam setup and user interface but has added in facial detection and recognition.
In this report, we share our test findings of Simplicam, matched up against Dropcam:
In particular, we examine Simplicam's key differentiator, facial detection and recognition.
- How often does the camera miss registered users?
- Does it mistake one user for the other and why?
- How will angle affect recognition?
- If you like Dropcam but don't want Google to store your home surveillance video, Simplicam has successfully copied all the key setup and user interface elements. It looks almost like an OEM.
- Simplicam has added a thumbnail / event card view (that Dropcam does not have) which makes it easier to review recorded video. However, the thumbnails are somewhat so it can be hard to see objects inside.
- Facial detection (i.e., identifying a face but not a specific person) worked ~90% of the time when people were moving towards the camera.
- Simplicam recognized enrolled subjects only about 30% of the time, despite being installed within manufacturer recommendations (~5.5' mounting height, ~5-6' to target).
- Given the camera's wide angle (107°) and short recognition range (5-6'), it was hard to get a good face shot without the person essentially walking up to the camera.
- Raising the camera to ~9' caused further reduction in facial recognition accuracy
- Frequent recognition issues such as recognizing two faces despite only one subject present, mistaking one user for another, or strangers as enrolled users
- One day recording plan is problematic, forcing users to review events quickly before they are overwritten.
Here are the key findings from our tests:
Technically, Simplicam's facial recognition is in beta, though available to all customers. We expect more UI features to be added but are skeptical about how much facial recognition performance can be improved, especially given the very wide FoV.
The Simplicam sells for $150 online [link no longer available], not including a recording plan. Recording plans start at $50/year for one day recording, $129 for 11 days recording, and $229 for 21 days. Recording plans are 40% off for cameras beyond the first one.
Given Simplicam's poor facial recognition accuracy, it should not be relied upon in home or commercial settings. With strangers recognized as users, and vice versa, along with simply missing detections, numerous false positives and failed activations are likely.
For technologically-inclined home users who are aware of these limitations and willing to deal with them, the Simplicam may be attractive, for the "wow" factor, as well as reduction in alerts as they are narrowed down by facial recognition.
As a Dropcam replacement, Simplicam is the closest we have found in consumer cameras, at ~$50 lower price point. However, their inexpensive recording plan is limited to a single day of storage, forcing users to either pay more than double ($129 for 11 days) for extended storage, or react to events quickly before video is overwritten.
The Simplicam is close in size to the Dropcam pro, though slightly larger in front diameter. It may be either desk/shelf mounted or wall mounted with the included wall clip.
Simplicam's facial recognition features allow users to enroll up to 10 people, a ~30 second process. Once enrolled, facial recognition events are shown in the Simplicam timeline and event view, with recognized faces overlayed on recorded video for comparison. Email and push alerts ("John Doe was detected at 1:38pm") may also be triggered on recogntion.
Simplicam recommends the camera be mounted about 5-6' high, and subjects be no more than 5-6' away.
Simplicam in Action
To better understand performance issues, watch this video:
In this sample image, our subject was detected properly, about 4-5' away from the camera, which is mounted at about 5.5' high. With him looking forward, and the scene well lit (~120 lux), his face is clear.
However, in the same scene, just a slight downward head tilt prevents recognition:
Raising the camera to ~9' high decreased recognition rates as well, to about 20-25%.
In addition to missed recognition, the camera regularly mistook one user for another, or an unregistered person as a user. This was especially common when both wore glasses.
Additionally, though less common, the camera regularly detected multiple people, despite only one being present. The overlay in the top left corner of the image below reflects this.