As the Police Officer working within the Forensic Video Unit, I was often asked to assist the Collision Investigation Team in dealing with CCTV and interpreting the data correctly. Since leaving the Service in 2015, I have witnessed good progress in the training afforded specifically to Forensic Collision Investigators (FCI) in dealing with video in a manner that preserves evidential integrity.

The two pillars of Forensic Video Analysis are Integrity and Authenticity.

Integrity examines if the item has changed since it was created, whereas Authenticity considers if the item is a true and accurate representation of that which it purports to be.

The analysis and consideration of these two elements is the key to using Digital Multimedia Evidence (DME) during an investigation.

A question may be, how fast is that vehicle travelling? Well, in order to answer that question, we must first understand what we are starting with.

Any conclusions on the timing of the video, or edges on an object could be wrong or questioned if there are unidentified issues with Integrity or Authenticity. We could spend the entire article on these two aspects as they really are so important but, for now, let us consider a few key points.

Video is easily changed through poor handling and a lack of understanding. Changes, unless controlled and documented, will affect both the timing and the visual data. An export of the original video must be used. Not a screen recording or a transcode from an internal data management system.  

Authenticity does not only cover ‘deepfakes’, it relates to original recordings as well. Perhaps a video has lens distortion on both sides causing everything to be curved. Is that authentic? No. 

Perhaps the video has been found to be transcoded (the creation of one digital file from another). Therefore the timing of the frames has changed. Is that Authentic? No.

Some of these issues can be corrected with restoration and enhancement techniques. Some, however, cannot be corrected but must be considered in any calculation and conclusion.

Now that we have assessed our video and know that it is an original native export from the Video Surveillance System (VSS), it is time to assess its suitability for Speed Estimation within Amped FIVE:

1. The vehicle must be travelling over a flat surface.
2. It must be possible to visualize where the bottom of the wheel has contact with the flat surface.
3. A rectangular reference object with known measurements must be visible on the same surface as the vehicle (an object can be placed on for this purpose).
4. A minimum of 2 video frames displaying the moving vehicle must be visible, although, with this minimum value, the uncertainty will be very large. More frames allow for a longer distance travelled and therefore a lower uncertainty value.
5. The frame timing of the video must be known and reliable.
6. The perspective of the video must not change throughout the range of frames selected.

If the video is suitable, an important initial processing stage is the correction of any optical distortion. Objects that are straight in real life, should be straight in the image.

This is where you will start to see the many benefits of an all-in-one solution like Amped FIVE. All decoding, analysis, restoration, enhancement, measurements and presentation can be achieved in a single project. It also avoids completing cross-tool validation, which must be completed if you have to use one application for one task, and then another for something different. 

The Undistort or Correct Fisheye filters can assist you in the task of image calibration.

Remember that any restoration and/or enhancement must be completed according to the image generation model. For instance, it is important to identify a requirement to correct the aspect ratio, as this should be applied before correcting lens distortion. Several image processing filters can also assist in visualizing the path of the vehicle, such as Levels or Exposure.

Once the video has been prepared for the next step, the Speed Estimation 2d filter can be selected.

The buttons in the filter guide you through the quick and simple workflow.

Grid: This allows the positioning of a perspective grid onto the image. Measurements can then be added to the width and height to allow for the measurement of distance.

Path: This allows for the vehicle path to be tracked. Multiple paths can be selected which allows for the speed of various vehicles to be calculated within the same range of frames. In tracking the vehicle, pixel uncertainty values can be set for the area in which the tyre makes contact with the road.

Time: This allows for the timing source to be set. Timing can be completed using the Timestamp, the MPEG Presentation Time Stamp (PTS), or the average FPS of that frame range. There is also the ability to include uncertainty in this value.

Speed: This is where all the data from the last three buttons get used, resulting in multiple different time, distance, and speed calculations.

Ruler: Allowing for multiple measurements in the scene, this can be used to validate the measurement of the vehicle path.

Export: With a huge amount of data now obtained, this button allows for it to be exported into a spreadsheet.

Now that we understand the workflow, let us take a visual look at each stage.

Setting the grid is a simple process of drawing the area on the video

The filter settings allow for the grid to be extended, to ensure the straight lines carry along a road edge for instance.

If the scene does not have any points usable to add a grid, then something can be placed into the scene to achieve this. Here is an example of placing a piece of flooring into the scene.

Amped FIVE allows for filters to be copied, so the grid can be measured with one video, and then used with the evidence video, as long as the camera has not changed position.

Plotting the path of the vehicle is as simple as ‘click and move on’. Each click of the mouse adds a point and then the frame is moved on to add the next point.

Remember, each point can have separate or linked uncertainty values.

In this image we can visualize the uncertainty boxes, to signify that the wheel is touching the floor somewhere in this 3x3 box.

We can now move on to timing. Timing is again a huge area that we could spend a lot of ‘time’ on.

FIVE allows for the full analysis, adjustment and presentation of timing information and with the ability of the Speed Estimation 2d filter to use any of the time references, all options are covered.

As the frame durations are important, it is often a good idea to display them on the image.

With the measurements set, the path of the vehicle tracked, and the timing reference added, we can move on and calculate the speed.

But it's not just the speed.

Multiple calculations can be presented at the same time. This again is one of the huge time savings of using FIVE. You are analysing, preparing, calculating, and presenting the findings at the same time.

All of this can be exported as a video, images, a frame-by-frame PDF, and of course raw data.

With this data, any graph can be created.

We worked with many FCIs in the development of this filter to make your life easier. Even with footage containing lens and perspective distortion, difficult grid placement, irregular timing and motion blur, the calculated speed ranges have been comparable with the much lengthier processes of calibrated speed runs and reverse projection.   

If you want to learn more and see the filter in action, head on over to YouTube to watch a recent webinar on this filter to learn how it links with some other filters within FIVE.

https://youtu.be/sxLLkYrE9Co

For current FIVE users who have gone through the basic Amped FIVE training, we also now have a dedicated training module on Speed Estimation. Click here for more details.

Finally, to request a trial or demo, reach us at:

https://ampedsoftware.com/contacts