The Sekonix cameras were the ones suggested by nvidia for use with the PX2. They came as either 120′ or 60′ field of view. We ordered 4 x 120′ to provide 360 vision around the vehicle which should allow 15′ overlap between each camera and 2x 60′ for a stereo pair that would be front facing. The street drone that Parkopedia ordered for the AVP project uses the same cameras but in addition to the stereo pair has 6x 120′, 1 at the front, one at the rear and 2 on each side. This configuration allows the camera overlap to be increased, which is no bad thing as the lens distortion makes the image less useful as we move further away from the optical axis of the camera. However, I was on a tight budget so I settled for the 4+2 configuration shown in the figure. After all, we could always add cameras at a later date. The PX2 actually supports up to 12 cameras via GMSL.
The cameras themselves are 1928 x 1208 resolution and supposedly waterproof (IP69K) as are the connectors so that’s the one less thing to worry about. However, as they are primarily aimed at automotive they are small (26x26x20.3) and do not have standard mounting points. A hunt of the web suggested Sekonix had some mounting brackets for several mounting options so I contacted them and they sent me the 3D CAD models. I printed one in pla on the printer (Hepestos 2) we have in the lab and it seemed ok. As shown in the figure it’s a basic cup designed to bolt to the back of the camera using 2 of the housing bolts, underneath it has a standard camera mount or at least space where a nut could be melted into the pla. My initial thought was that pla would be too brittle, especially given the lack of suspension on the Twizy. So I investigated getting some machined in alu. It was cheaper to outsource this than do it in-house (which is sad) but it was still going to Cost $100 per mount and mounting it on a tripod mount also filled me with fear about subsequently trying to calibrate it. So I put the decision on hold until I’d at least got a better idea of what the roof structure for the lidar would look like.
Once the extruded aluminium and lidar were mounted on the roof it made sense to mount the cameras to that. Originally we didn’t think it was forward enough on the vehicle to mount the front facing cameras and so I resigned myself to probably having to mount them directly to the roof itself above the windscreen. But some tests with the cameras themselves suggested that as long as they were mounted high on the lidar tower we were only going to loose around .5 meter (at most) of floor coverage in front of the vehicle. These tests also suggested that the side and back cameras should have an inclination of 45′, the front camera a 0′ inclination and the two stereo cameras 9′ inclination. The only issue is mounting the stereo cameras high on the lidar tower means a narrow stereo baseline. Putting them nearer the side cameras would give a wide baseline but much lower mounting position which means less coverage immediately in front of the car. I decided to go with the former and hopefully push the baseline out slightly by mounting either side of the tower. It can always be changed later if needs be.
The Motedis website also provides CAD models for the extruded aluminium which helped in building the mounts.
I used 123D shape to add brackets to the Sekonix camera mount but I had some problems creating a single solid model so once finished they were remeshed in mesh mixer, making them solid before exporting to stl and then to gcode via cura.
Photos or the model, the final print and it’s placement on the car are below along with the stl files.
You can download the .stl files from Thingiverse here. I had no problems printing at all, didn’t even need to recalibrate the printer. Then I mounted them to the car. Unfortunately, on only one of the models did I remembered to put mounting holes (the last one I did), the rest I just used a 5mm drill bit. I normally print at 20% infill but all of these were printed in solid pla (100% infill) to try and make them as strong as possible, as such drilling them was both easy and produced very clean holes.
Oh and I stuck an amber flashing light on, because I thought it would be cool 🙂