What main factors/features explain the high price of most industrial computer vision hardware?

Question

I am a student who is currently working on a computer science project that will require soon computer vision and more specifically stereoscopy (for close depth detection). I am now looking for a great camera to do the job and I found several interesting options:

1- A custom built set of two cheap cameras (i.e. webcam);

2- The old, classic, economic and proven Kinect;

3- Specialized stereo sensors.

I found a couple months ago this sensor: https://duo3d.com/product/duo-mini-lv1

I tought it was interesting because it is small, stereoscopic and brand new (encouraging a fresh USA company). However, if we take apart the additional APIs that come with it, I just don't understand why you would buy this when a Kinect (or "cheap" cameras) are at least 4-5 times less expensive and still have great if not better specifications.

Same applies for this one: http://www.ptgrey.com/bumblebee2-firewire-stereo-vision-camera-systems

Can someone please explain to me why I would need such a device and if not, why they are needed for some?

Answer 1

So the Kinect uses a set of IR points to determine depth, not true stereoscopy. the DUO M and the Bumbee both use two cameras to achieve this, I am not familiar enough with them to give a good comparison between them and the Kinect but I am familiar with a new product that I have been using for a little bit called the ZED It has a detection range of approximately 30M vs the Kinect's 1.2–3.5 m, the ZED can operate outdoors, the Kinect cannot due to it's IR based ranging. However the difference is that the ZED needs significant processing power to do the image processing at a reasonable rate, it uses the GPU functions in OpenCV to accomplish this, I am using it with an nvidia K1 board on a Quadcopter. The advantages I outlined with the ZED hold true for most dual camera stereoscopic setups.

Answer 2

As Mark Omo mentioned, the Kinect is a fundamentally different technology than a simple stereo pair. There are also other technologies such as a "standard" time-of-flight cameras like the Swiss Ranger and PMD, and infrared stereo with projected texture of the Intel RealSense.

I believe each technology has its own strengths and weaknesses. For example, standard stereo cameras typically can't get depth information on a single-color featureless plane (like a large wall). But time-of-flight cameras have no issue with that. However, time of flight cameras will have trouble with materials that absorb light (like very dark objects) or objects with other interesting optical properties (like snow). Many optical based sensors will have trouble with clear and colored glass, so you may need to completely change domains and use a 3D ultrasonic sensor.

Furthermore, all sensors have min and max ranges, however active sensors (like the Kinect and time-of-flight cameras) will typically have smaller max ranges due to power requirements. Whereas increasing the max range of a stereo pair is as easy as increasing the baseline.

So you should pick your sensor based on your target application. There is no one-size-fits-all 3D sensor.

Answer 3

As others mentioned the Kinect is a different technology which has different trade offs. Stereo performs better in sunlight than a Kinect and it is easier to adapt the range, particularly if longer range is desired.

The reason that the stereo modules are more expensive than a couple of web cams is because they are not the same thing. The technology itself isn't that much more expensive, but it doesn't have the pricing from large commodity volumes. For the stereo modules, they typically have global shutter cameras. Almost all web cams are rolling shutter cameras. The difference is that global shutter cameras expose every pixel at the same time. Rolling shutter cameras capture each pixel line at different points in time. Web cams tend to be very bad with motion, particular of the camera. This doesn't work well for stereo if things are moving for many applications. Typically people want to get a point cloud or similar from the stereo but if it is captured while the things are moving it won't be consistent. Stereo modules also have synchronization of the camera captures. If things are moving, it is important that both cameras capture the images at the same time. In a lot of stereo modules, this is done by sending a pulse to both cameras to start each capture so that they happen at the same time. Another factor is that for stereo to work, the cameras need to be in a known position relative to each other. Typically this means that there is a rigid mount between the cameras that keeps the relative position fixed. Then this relative position is calibrated. For some stereo modules this is done at the factory. This can also be done pretty easily with open source software tools, but if the cameras move relative to each other it has to be redone. Finally, some stereo modules also have hardware acceleration to do the stereo calculations on the stereo module. The cheapest alternative is to build your own with 2 global shutter cameras with appropriate lenses that support external triggering and put into your own rigid mount. It isn't so easy to find the appropriate cheap parts to do this for less than the cost of the cheaper stereo modules. If nothing is moving when the stereo ranging is desired, than cheaper technologies like web cams become viable again, but that is why the stereo modules cost more than a couple web cams (along with the low volumes needing to cover development expenses).