cjones20
New Member
My intention for this post isn't to spam anyone. I'm not selling anything (no urls). I'm doing some market research for a new company I've started and want to get some insight from dashcam users.
I'm designing a dashcam with stereo vision (two forward facing cameras), IMU (accelerometer, gyro, compass), GPS, and it connects to the OBDII connector on your car to record steering angle and speed. The data collected will be used to train computer vision models to help self driving car companies do a better job at predicting what paths vehicles and pedestrians will take, where lanes are, stop signs, stop lights, etc.
The basic idea is that you buy a dashcam (maybe $400, very high end components) and as data is uploaded to the cloud that we can use we credit back the money until maybe after 12 months we've credited back the entire amount. The data would be anonymized. I have no use for any identifying information about the driver. The reason it's not just free to begin with is we want to make sure people don't order them and either not use them, not upload the data, or use them for parts.
The cameras I'm looking at are what's called global shutter (all pixels are recorded at once). Most dashcams use a camera called rolling shutter (each row of pixels is recorded separately). The rolling shutter combined with a slow shutter speed results in motion blur. When you pause the video a picture of a car driving by looks blurred. This is no good for computer vision, we need sharp pictures of pedestrians and cars to accurately classify them.
Are there any particular features in a camera that are must have? Automatic generation of maps of where you've been? Ability to share videos straight to Youtube?
I'm designing a dashcam with stereo vision (two forward facing cameras), IMU (accelerometer, gyro, compass), GPS, and it connects to the OBDII connector on your car to record steering angle and speed. The data collected will be used to train computer vision models to help self driving car companies do a better job at predicting what paths vehicles and pedestrians will take, where lanes are, stop signs, stop lights, etc.
The basic idea is that you buy a dashcam (maybe $400, very high end components) and as data is uploaded to the cloud that we can use we credit back the money until maybe after 12 months we've credited back the entire amount. The data would be anonymized. I have no use for any identifying information about the driver. The reason it's not just free to begin with is we want to make sure people don't order them and either not use them, not upload the data, or use them for parts.
The cameras I'm looking at are what's called global shutter (all pixels are recorded at once). Most dashcams use a camera called rolling shutter (each row of pixels is recorded separately). The rolling shutter combined with a slow shutter speed results in motion blur. When you pause the video a picture of a car driving by looks blurred. This is no good for computer vision, we need sharp pictures of pedestrians and cars to accurately classify them.
Are there any particular features in a camera that are must have? Automatic generation of maps of where you've been? Ability to share videos straight to Youtube?