Featured Post

Welcome to the Forensic Multimedia Analysis blog (formerly the Forensic Photoshop blog). With the latest developments in the analysis of m...

Friday, June 21, 2019

Practical Field Photogrammetry

Another successful Photogrammetry class is in the books. I'd like to share a few thoughts on the experience, as well as an example we worked on in one of the field exercises.

The old adage, "a poor workman blames his tools," is only part right. If you don't have the right tools to begin with, you'll be up against it. In the justice system, analysts are often stuck with what they've been provided - by the lowest bidder. This is something I realized years ago in that crime factory of a city that I used to work in. I needed a kit that was economical, rugged, portable, and easy to work with. It needed to be economical because the expense of kitting myself out was coming out of my own pocket.

As regards measurement exercises, I've seen many height charts / resolution charts on offer at trade shows and on-line. The ones in my price range were often printed on ink jet and when mounted, mounted on cardboard with a rigid frame. They were bulky and awkward. They maybe would last a month of regular use.

Having a background in art, working as a commercial artist and having run art departments in my youth, I decided to design and manufacture my own resolution chart.


Beginning with the end in mind, I decided to not go with the standard checkerboard style. I didn't want the black boxes to touch each other. At a distance, and given the massive compression on most surveillance systems, I wanted to avoid the bleed-over that often happens with the checkerboard designs. Take this from the Amped FIVE sample folder, as an example of the bleed-over.


This image is a close up, and in the camera is less than 12 feet from the chart. Still, you can see the problems of trying to establish the corners for an "eyeball" nominal resolution calculation.

The types of enquiries we often face deal with questions of license / registration plates at distances of 50'-250'. Take this example from class.


The camera is mounted on a light pole about 100' from the target, elevated about 12' from the road. I've processed it for display here in FIVE. The system is on the higher end of the generic black-box Chinese DVR spectrum and is well maintained. The owner of the system is a friend and the strip mall isn't that popular, so it's a nice place to have a field day.

My resolution chart is commercially printed on a PVC board. It's durable, water proof, and can stand propped up against the back of the car. Thus, I don't need to carry a stand or easel. It fits nicely behind the driver's seat of my truck.


Cropped in on the back of the car and the chart, this is the view at 100'. The left-most blocks are 2", the middle squares are 1", and the right-most squares are 1/2". There's very little bleed or compression artifacts around the 2" squares.


There are two stickers on the rear window. What's the diameter of the top sticker? I've got a good look at the 2" squares, so I can use those as my references. I can also use them to calculate the nominal resolution. If I use the Ruler in FIVE to measure the pixel length of the 2" squares, I get 4px.  Thus, the nominal resolution around the rear of the car is 2px/in.


Moving on to measurements of the items of interest on the vehicle, accurately selecting the X and Y axis in FIVE takes just a minute. The results are displayed in seconds. Things like stickers help to "individualize" the car and differentiate it from the other Honda Fits in the general area. This one has a 6" diameter round sticker on the center left of the rear hatch's window, above another sticker.


We can check our work by measuring an object of a known length.


US license plates are 12" long. This car appears to have a license plate frame. We measured from the outside of the frame on both sides. The results are again displayed. Assuming a bit of error (Measure 2D in FIVE does not compute error as 2D measure is simple planar geometry), at 100' from the camera and with compression, that's not a bad result - 1/2" frame width on both sides is not out of the realm of possibility.

What makes this possible? An easy to use resolution chart.

This is a simple example. But the technique is the same for your basic reverse projection photogrammetry experiment. Make a recording of the scene with the chart placed appropriately, then overlay the two recordings (that's the reverse projection part) in order to facilitate the measurement (that's the photogrammetry part). In this case, everything is roughly on the same plane, so I used FIVE's measure 2D. FIVE has 1D, 2D, and 3D measurement capabilities.

The last part of this exercise was the validation of the results. Validating results, in each and every case, is a must. There is no research out there that can be utilized that "globally" validates reverse projection photogrammetry. Every case is unique - different camera, different lens, different DVR, etc. Thus, you validate. The few text books out there that mention RP advise validating each case. In the SWGDE doc on photogrammetry (link), they advise the analyst to calculate / determine "... the accuracy and precision of this result." That's just another way of saying, "validate your results."

Keep in mind that if you're using a mixed-methods approach (laser scan point cloud + still CCTV frame), that method has never been validated outside of some analyst's casework. Remember, getting evidence and testimony accepted in one venue is not validation. So please, do your validations.

There's one more Photogrammetry class scheduled for 2019. You can check out the calendar of classes on offer here. This course tends to fill fast as there are no "non-vendor specific" courses on the subject of photogrammetry available in North America. It's a very intense week as we dive deep into the science and practice. It's a very tiring class to teach. It's a lot of information to transfer in a week's time. But, I enjoy it.

When students take this course, they've usually been through the LEVA levels and the various tool-specific training sessions. They're looking for more training, or topic-specific training. They're usually well kitted-out for the basic stuff. Thus, what do you get for the person who has everything? A squeeze ball and a few branded pens? No. Students get the digital printing file of the height chart and instructions on contacting the commercial printer we use. The cost of printing and shipping is usually less than the cost you'll find in the forensic magazines or at trade shows for their charts. If you can't make the class and would like the file, a small stipend is all it takes. If you'd like me to arrange the printing and have it shipped to you, the small stipend + the printing / shipping costs is still reasonable (I don't resell the printing/shipping). Either way, just send me a note and we'll get it set up for you.

Until then, have a great summer ...

No comments: