I would not expect, that fix to those ghost images will be released soon by Novatek, as it need a lot of work.
HDR used here is capturing two exposure images at different exposure levels and combining those to single image. Unfortunately it is done in frame after frame mode without motion detection and motion compensation algorithms. Algorithms that will use some CPU resources.
HDR at 30fps means 16,7ms delay between two frames (long and short exposure ones).
At 36km/h we have 10m/s = 1000cm / 1000ms
1ms delay is 1cm and we have over 16 cm ghosts on those images in side areas.
Better would be use of DOL-HDR build in Sony sensor that read out each pixel twice in succession and combine those in the sensor to form an HDR data line. I believe that this sensor uses even the second-generation DOL with virtual channel technology (feeding CPU with two image streams one from short exposure and second from long exposure). Still some some motion detection and motion compensation algorithms need to be implemented but will have easier job to do.
View attachment 704
With implementation of DOL-HDR in two streams mode, the delay (between long and short exposure frames) is greatly reduced (I would bet between 50x to 100x = just 1,6mm of ghosts).
Introducing Active Radiance Map of created HDR images with feedback from tone mapping algorithm would allow to actively change exposure time on frame basis and improve local contrast. (now when entering or leaving tunel the reaction is too slow and final image to dark). That why many users see -EV in settings as a must.
Introducing ROI (Region of interest) in a belt from left to right with height of around half of the sensor lines would allow concentrate CPU resources for this most important area.
At night detecting two light sources (car headlights) in ROI would allow to concentrate even more on area between them.
Other (than in software) possibilities are in hardware and particularity in IMX464 sensor. At twice better base sensitivity than IMX335 and almost five times higher usable one it is something for A119 V4 but its size (9mm from 6,5mm) means new approach for DVR build and higher price.
I am really looking forward for good news and for implementation of as many thing as possible as soon as possible.
