It seems impossible to read the license plate on the semi-truck on the left, but think it might be possible to get a read of the mudflap. I tried this in adobe photoshop to no avail, any help would be greatly appreciated! Looking for something to identify the truck that fled the scene of the accident.
+ Reply to Thread
Results 1 to 18 of 18
Last edited by Car_Accident; 6th Dec 2022 at 22:05.
Besides me not really seeing a distinct mudflap, due to the glare of the taillights, being realistic, you should count the pixels that would make up the region of interest. The absolute bare minimum that can create a recognizable letter/digit is probably 5x9, and more likely a reasonable minimum is 4x or 9x that (10x18 or 15x27), for EACH CHARACTER. If your image doesn't accommodate enough for that, there is not enough resolving power, even using multiple/various convolution/sharpening methods.
Well that is discouraging! So basically there is no way to identify the semi-truck!? I was hoping the mudflap or even a symbol/logo on the back the truck would shed more light to the situation.
I am not sure I follow regarding the pixels and would not even be sure how to figure that out. Any chance you can determine what the pixels are on the video? I would even be open to paying an expert for the analysis so long as it can produce some form of image that would indicate what transpired.
Last edited by Alwyn; 7th Dec 2022 at 03:13. Reason: Not an appropriate comment for this forum.
You can find dozens of videos on YouTube showing how dashcams can almost never resolve a license plate. At night, pretty much 100% of them fail. Resolution is typically not the main problem, but instead their failures are caused by sensor dynamics (ability to deal with huge contrast and brightness differences); motion blur (things are traveling fast); lousy lenses (you can have infinite sensor resolution by if you shoot the video through a Cracker Jack plastic magnifier lens, nothing will be sharp), and more.
In the case of your video, there is nothing there at all. Sorry, no hope.
Even the mudflap? I feel like I can almost make it out
So, I have done forensic work before. There is a big difference in the spectrum between (am using my own terms here):
1. Recognition/Noticing that SOMETHING exists on the screen
2. Identification that the SOMETHING is a particular OBJECT
3. Resolving that OBJECT as something potentially LEGIBLE
4. Evaluating/"Reading" that LEGIBLE object as a specific, unique and distictly identifiable DATA.
5. Verifying that DATA, along with other supporting facts and features, as being reliable EVIDENCE.
Here is an example of a possible text you would have a 400% better time recognizing than what currently exists on the above image (exact same scale as the clip).
[Attachment 68048 - Click to enlarge]
It is at the size that it might appear on a mudflap where in reality it might be noticeable/recognizable as text to the human eye. I overlaid in on what might be a place where one of the existing mudflaps would be (using Multiply function), and I used a high contrast (Black on White) text. However, given the motion, I also added the absolute smallest amount of gaussian and motion blur, to simulate what it should be expected to look like in real life.
TRUCK # (serial number)"
Can you read the serial number? Can you even make out the "Van Lines" portion?
That is the scale we are talking about.
You are wanting something like #4 or #5 above, but most images of this nature will only provide #2 or #3.
Last edited by Cornucopia; 7th Dec 2022 at 14:50.
I have read several articles and watched several YouTube videos in just the past week about how virtually all of the dashcams can't provide license plate numbers except under the most ideal circumstances. I got a Viofo dashcam for Christmas two years ago and plug it in every time I drive (never bothered to mount it permanently). I can easily read the plate of a car stopped in front of me at a light in broad daylight, with the sun overhead, but driving into the sun; driving at night; trying to get the plate of a car approaching on a two-lane highway, and other similar situations make getting the plate highly problematical.
So my personal experience confirms what these articles are saying.
Good solutions do exist, but generally at a price point of over $1,000 per camera. I have the specs of such a system that is used by one of our local towns and would be happy to share them. This town has successfully identified, apprehended, and convicted several bad guys. See the article on the bottom of the first page of last week's local newspaper for more information:
License plate is an impossibility, I get that. But what about the mudflap or anything on the truck that would help identify the other vehicle. Anything would be a great help!
Here's an even bigger magnification. Don't you understand that there has to be some hint of an image in order to be able to make something appear? My point in posting that "enhance" video in my previous post was to show you that Hollywood movies have influenced you to believe that you can magically make a useful image appear from a mess of pixels like you see below.
This is 100.000% false.
Last edited by johnmeyer; 8th Dec 2022 at 13:08.
Even if you have a clear picture of the mudflap you wouldn't want to pay thousands of dollars for an investigator to track down certain manufacturer's product and when or where it was sold and to whom it was sold to, Unless there is a fatality which requires the law enforcement to go through that pain for you. From the scene in the video I don't even see any property damage, So let the grudge go and try to drive carefully next time even if it wasn't your fault, and if you ever need to get some information get it right at the scene with your eyes, there nothing better than the human eye.
The damage on the vehicle was pretty severe even though the video doesn't seem like it is. No sound makes it seems a lot less impactful than what really happened. The entire front cab is destroyed and including the cargo we are looking at over $200,000 in damages without the injury that occurred to our driver. It did not result in a death but our insurance only covers about 1/4 of the damage and our driver was taken to the hospital. Thankfully, the driver with the dashcam did clearly see who the other party was (a large carrier), but having video evidence would undeniably allow us to identify the other party.
I assume you've already sent the original (always keeping a copy for yourself) to the authorities / insurance carrier. The cops often have some specialized forensic software, although my experience is that it isn't all that much better at restoration, but is better at machine recognition/categorization. You could try tracking the rig in Adobe AfterEffects (or similar), using the tracking to create an extended non-moving/stabilized shot, and then run a median filter on the combination of frames making up that stabilized shot. This kind of noise reduction can assist with local contrast enhancement. But you still have the overall issues of lack of resolution (of the identifying data regions) and camera/sensor lack of dynamic range adaptation, as has already been mentioned.
Beyond that, I don't think there is much you can do with this footage.
Scott, I went through one frame at a time and there is not one single frame that reveals any detail whatsoever. In order to get averaging across frames to work, there has to be some semblance of structure somewhere, on at least two frames. This doesn't exist in this sample.
I'm not surprised, but I was trying to give it the benefit of the doubt.
I appreciate all the feedback, it seems like the video will not help provide any indication of the other party :/