Concerning video evidence of close passes.

Many police forces are now accepting online submission of reports from cyclists (and motorists) of bad driving – particularly close passes – with video evidence. Although there have been online reporting systems for some time, such as the Metropolitan Police’s RoadSafe London, the current trend in providing this for cyclists with the expectation that reports would lead to prosecutions was started by West Midlands Police as part of their acclaimed Operation Close Pass initiative. So, I was very pleased when Greater Manchester Police Traffic Division approached me back in February to participate in trials of such a system here as part of their Operation Considerate. This reporting procedure is now in place and functioning and, from my point of view, is going reasonably well.

However, one thing that is always in the back of my mind when making these submissions is that, as far as I am aware, this approach to dealing with bad driving has not yet been tested in court here. I am not wholly confident that the video evidence in a lot of cases will be sufficient to convince magistrates or jurors that the driver was guilty of the offence of careless and inconsiderate driving. Both magistrates and jurors are lay people as far as the interpretation of such evidence is concerned. I’m confident that, after looking at these videos for years and doing quite a bit of work on calibrating my own front and rear cameras, I can judge fairly accurately the distance between a car and a bicycle in most videos, provided the relevant visual cues are present (more about that later). I am equally sure, though, that if I had been asked to do that 7 years ago, I would not be at all confident, and that is the position that most magistrates and jurors would be in.

The problem is that to judge these distances scientifically can be quite difficult and fraught with many potential sources of error. There is an excellent write-up of how you could do this, if the conditions were right, by Bez in an article on his Beyond the kerb blog, which is well-worth reading for people who are interested. However, most people are not going to be able to do this as it requires some specialist knowledge. The usual method in cases where specialist knowledge is needed to interpret evidence is to get a specialist to do the analysis and then present the results as evidence in court. However I am not sure of the extent to which the police are likely to do that in a close pass case, and I’m not completely confident that they’d make a good job of it given the example very crude evidence that was presented on stopping distance in the recent Charlie Alliston case.

A crude experiment

After one of my video submissions to Operation Considerate was bounced back on the basis that it did not contain enough information to estimate the passing distance convincingly, and having discussed that case briefly with a couple of other people, I decided to set up an fairly crude experiment using a Twitter poll to see how a number of other people would judge the distance in one specific picture. Here is the result:The correct answer here was 1.0 metres, as 55% of the respondents guessed. I was quite intrigued by the other answers, though. I was initially surprised to see so many saying 0.5 metres, and also by a few (in fact 4) people guessing 2.0 metres. I was expecting more people to guess 1.5 metres, but I  do think there may be considerable bias in here as I will discuss below.

Note that I think the crudity of this exercise is acceptable because I’m trying to get a handle on people’s perceptions rather than trying prove some consequence of physics and, anyway, I’m not submitting it as testimony in a court of law. However, it is worth pointing out there are caveats and potential sources of bias in this exercise:

  • There are visual cues in the picture that wouldn’t be present when travelling along a road; particularly the kerb stones across the far end of the parking space. (It wasn’t practical to do this quick and dirty exercise out on an open road.) This may well increase the number of people choosing the correct answer.
  • Given the nature of my Twitter channel, responses are likely to be primarily from cyclists, and even from cyclists who are used to judging distances in images and videos like these. Again, this is likely to result in a greater proportion of correct responses, but I suspect it may also be the reason for the large number of people under-estimating the distance due to unconscious bias. Had the responses been mostly from non-cycling drivers, I can imagine there might be far more people over-estimating the distance, like the 9% who selected 1.5 metres or more here.

Some sample shots

The picture in the twitter image comes from a set that I took in order to get a handle on how the gap looks for different distances using my camera, a Contour Roam. I parked the car in a space, then drew chalk marks on the road in a line perpendicular to the rear wheel, at intervals of 0.5 metres, starting at 0.3m, which is half the width of my handlebars. The image below shows one of my bikes at the zero position.

You can see the marks on the ground. Note that with a very wide angle lens like this one (170 degrees), there is significant barrel distortion, which makes it look as though everything at the top of the picture is falling away to the sides and the chalk marks appear to be in anything but a straight line. This is not so much a distortion as just a consequence of the fact that, for example, at the edges of the picture the camera is looking almost sideways to the left and right. However, it does imply that distances might not be all they seems to be, and to calculate the correct distances can be quite complicated.

Having said that, let’s have a look at how the images appear at the different distances. From top-left to bottom right, these are: 0.5m, 1,0m, 1.5m, and 2.0m:

(Click on an image to see it at full-size.) I have selected these approximately to get the near-side rear wheel just into the right of the shot, though that becomes increasingly difficult to do at the larger distances.

The thing that seems quite striking to me on looking at these images together is that 0.5 metres difference in passing distance makes a huge difference to the appearance of the shot. Even with the possible distortions, head position, etc., I think it should be immediately apparent in most video images from the same camera whether the car was 1.5 metres away or 1.0 metres away; 1.5m being the recommended gap at 30 miles-per-hour.  At higher speeds, such as 40 mph or more, the image should be looking more like the one on the bottom right; indeed, at 50 mph I would say even that one should be considered too close.

An interesting point to note about the barrel distortion that I mentioned earlier is that, if we look at the 2.0m image, the car looks as though it is well past the front the bicycle. However, the chalk marks on the road reveal that not to be the case. In fact the car’s rear wheel is just very slightly in front of the front wheel of the bicycle. Remember, at the far right hand edge, the camera is looking almost sideways!

All of this makes me far more confident of the value of the video evidence I am providing to Greater Manchester Police under Operation Considerate. That’s not to say there wouldn’t be problems in a real case in court; after all, a defence counsel only has to say sufficient to put “reasonable doubt” into the minds of a jury to secure an acquittal, but that can only really be tested in court.

But what about different cameras?

As I said above, these images make me a lot more confident that such video evidence is certainly of use, but it does not address the problem of how jurors and magistrates who are not familiar with my camera would interpret my images, nor how they might interpret images from a range of different cameras. One of the key differences between the cycle cameras that can be typically bought is the field of view. As I said, my own helmet camera has a claimed field of 170 degrees (though I haven’t actually tested this yet). I have had another camera in the past with a 120 degree field of view, and Cycliq claim that their Fly12 camera (which is the one police forces are tending to use for close pass operations) has a 130 degree field of view, and I believe their Fly6 rear camera is 100 degrees.

As a quick test in relation to this issue, I cropped my images down to simulate using a camera with roughly 120 degrees field-of-view. The following images show the original still from the video at 1.0m next to the cropped version of the same image, with (below) a similarly cropped image taken at a position further back to get the rear wheel of the car into the edge of the frame.

I have to say that I find neither of the cropped (narrower field of view) images as convincing as my original very wide angle image. One of the reasons for this is that there is nothing in the frame that provides a reference to the position of the bicycle.

Even though we are only looking at still images here, I think it is absolutely crucial in the interpretation of these videos that there is part of the bicycle in the frame to provide a reference. In my case, I very much favour the wide field-of-view of my camera in 960p format (rather than 720p), which allows me to get the handlebars in the bottom of the frame and still see the sky above the horizon at the top of the frame. My camera can also record in 1080p, but I don’t use that mode purely because the field of view is much narrower.

The other issue I think is important is unfamiliarity, or lack of a set of sample images against which to make a comparison. In the section above showing the appearance of different passing distances, it seems clear that such a set of reference images from the same camera would be extremely helpful to anyone being asked to make a judgement from a single case in a video or still-frame. I would go so far as to suggest that the provision of such a reference set should be a standard part of the evidence brought to a case by the police.

Adding a raster to images

One of the things that I have done for a long time myself is to superimpose a raster onto a still image taken from a video to help in the estimation of passing distances. I made a video showing an example of how I’ve done this for another blog article about a specific incident:

I hope this is self-explanatory, but basically it involves setting up a tape measure on the floor, placing some markers (canes) at 0.5m intervals, then getting on the bike with my camera mounted and moving my head around to collect lots of images for different head positions. These are then turned into a set of rasters from which an appropriate one can be selected and superimposed on a still image in any particular case.

I have never been completely confident about the accuracy of this technique, though I believe, after using it for a while and making comparisons between images at different head positions, that it should be better than + or – 20cm. So, one of the things I have done with the set of images I showed above is to put my rasters onto the images and see how well they match. The results are here:

Again, these images are for 0.5m, 1.0m, and 1.5 m respectively, and the markings on the raster are at 0.5, 1.0, 1.5 and 2.0m from the front tyre of the bicycle. Allowing for the 30cm lateral distance from the bicycle tyre to the end of the handlebars, this gives a pretty accurate indication of the distance, despite the fact that these rasters were done using a different bicycle to the one I used to create the pictures, and the head position in the image from which I created the rasters was not identical to the head position in these images. Having said that, I am under no illusion that this analysis would be admissible as evidence in court; although I have to say, it is no poorer in quality than the evidence of stopping distances presented by the police in the Charlie Alliston case, to which I referred at the beginning of this article, and probably even somewhat better.


This has not been a particularly scientific analysis, but I would assert that there are nevertheless some specific lesson that can be drawn from it:

  • It is possible for even lay people to judge distances in such video evidence, but the conditions need to be right.
  • One of the conditions required is that there is a part of the bicycle visible in the frame as a reference to its position.
  • In general, very wide angle videos (~170 degrees) are preferable to narrower ones.
  • The judgement of distance is likely to be far more reliable if a set of reference images from the same camera is provided to show the appearance of images at different passing distances.
  • It is possible to use a raster superimposed on stills taken from videos to indicate with accuracy the passing distance of an overtake.

I intend to write a second article on the criteria that might be used by the police in triage to determine whether to pursue a case and in what way; how that might relate to quality of evidence; and how such criteria might relate to the specific objectives of the close pass operation itself.

This entry was posted in Article, OpConsiderate, Police, Road Justice. Bookmark the permalink.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.