VoxelFlow consists of three event cameras and a continuous laser scanner, whose simultaneous observations use triangulation to generate a 3D image within a matter of nanoseconds. The system is able to detect, track and trace that 3D image, in addition to specifying velocity, speed, direction and position. Unmatched in its field, VoxelFlow will generate the best way to perceive the motion of 3D objects in space, creating a new way of seeing within the automotive industry. The system cuts the reaction distance from six meters to twelve centimeters and will be 50 times faster than other existing technology. By significantly cutting down response time and generating an authentic view of the world around the vehicle, VoxelFlow will help ensure the safety of all those on the roadways.
At 250,000 voxels per second, VoxelFlow is already matching competitors in point density. Where current camera/laser technology is subject to delays in latency or deficiencies in image quality, VoxelFlow creates high-resolution, low latency images, moving beyond pixels to reimagine the world in voxels. Sensor technology designed specifically with AD/ADAS vehicles in mind, VoxelFlow’s low latency caters to object detection within 30-40 meters, it shortens the braking distance for the vehicle. Further, the hybrid event-based camera and laser approach, allows VoxelFlow to work across all forms of weather and during every time of day.
Terranet publishes a video about its event-based camera scanner technology:
Event cameras have a latency proportional to the light level: yet another laser-based ADAS sensor that will burn our eyes as long as the automotive world prevents serious studies on the topic.
ReplyDeleteWhat if an event is triggered by the scene in the meantime, is the laser impact still detected/tracked ? Yet another system that will probably work with the NCAP -sparse- scenarios, but not in real life.
It's a cool idea, but not for automotive
Who is the sensor provider ?
ReplyDeleteAny scanning Lidar can stream with even lower latency = almost any Lidar on the market
ReplyDelete+ its difficult to get enough light onto the image sensor with small pixels with eye safe laser power
+ multi device interference
The potential benefit of event based cameras in the ADAS space is tremendous, now that 5um pixel are available at Prohesee/Sony, Samsung and possibly Celepixel/Omnivision the gates are full open for startups such as Terranet. I am actually very surprised there is not more of them out of stealth yet.
ReplyDeleteIt's not that surprised to me. Event cameras or similar things bearing other names in earlier days have been round for quote some time. But it struggles to find killer applications. A recent research posted here by Vladimir Koifman (http://image-sensors-world.blogspot.com/2021/02/do-event-based-imagers-have-advantage.html) still shows no obvious advantages comparing two commercially available event cameras to two commercial fast frame rate global shutter cameras.
DeleteOne thing I don't understand is that why an event camera can't be sold at very cheap price. My naive understanding is that an event camera just uses more transistors per pixel (in a way similar to many so-called smart CIS with multiple capacitors/memories and associated control gates). Adding complexity to pixels makes them bigger thus should just modestly increase the cost scaled by the die area. It's not a really a chicken-egg problem as they say. Because the CIS foundry ecosystem is wide available there and these mature node processes NRE is not that expensive. I don't see the obstacles of selling an event sensor at a similar price as some small volume high end CIS sensors.
What is the usual price you know or have heard of an event camera? True, a sensor price could simply scale with Si area, but not everything sells like a commodity product. There are other market forces like R&D needed to commercialize a product, barrier to entry in the segment, availability of other options, etc.
DeleteIt seems that the consensus on the article you are citing is that the article is of bad quality.
Delete