Angle Measurement Tool

An angle is the easiest thing in the world to measure on a picture. Click a corner, click a point on each side, and a number appears with two decimals. Nothing needs calibrating, because an angle does not care how big the picture is.

That ease is the trap. The arithmetic is exact, and it is exact about the picture. Whether the picture tells the truth about a knee, a rafter or a bracket is a separate question, settled before the shutter opened. The argument of this essay is that the second question deserves most of the attention, and that a measuring tool earns trust by saying so on its face.

What an angle is, and the units for it

The school definition is better than it sounds. The Common Core mathematics standards describe an angle as the shape formed wherever two rays share a common endpoint, and they measure it against a circle centered on that endpoint: the fraction of the circular arc between the two rays. An angle that turns through 1/360 of a circle is called a one-degree angle. Everything else is counting.

The definition explains why scale drops out. Draw the circle twice as large and the arc between the rays doubles, but so does the whole circumference, and the fraction is unchanged. A photograph printed on a postcard and the same photograph on a billboard hold the same angles.

The degree is not part of the International System of Units, but the National Institute of Standards and Technology lists it among the units accepted for use with it, and defines it through the radian: one degree is pi divided by 180 radians. The radian itself is a length divided by a length, meters per meter, which is why the same guide calls it a special name for the number one. Half a turn is pi radians, 180 degrees.

Two more units survive in particular trades. The gon, also called the grade or gradian, is exactly 0.9 of a degree, so a right angle is 100 of them; it came out of France in the late eighteenth century alongside the metric system and is still used mainly in surveying, especially in Europe. The revolution, or turn, is the whole circle: 6.283185 radians in the federal conversion table.

Minutes, seconds, and an old way of counting

Below the degree, the customary subdivision is by sixties. One minute of angle is one sixtieth of a degree and one second is one sixtieth of a minute, both written into the same federal guide. An angle of 36.87 degrees becomes 36 degrees, 52 minutes and 12 seconds.

The pattern is the one on a clock face, and it is very old. Historians of mathematics at the University of St Andrews trace the base of sixty to the Sumerians and Akkadians, from whom the Babylonians inherited it, and note that the division of the hour into 60 minutes and the minute into 60 seconds has survived for 4000 years. They are careful about why sixty was chosen: several theories exist and none is settled.

Seconds of angle sound absurdly fine until they are laid on the ground. The United States Geological Survey gives one degree of latitude as about 69 miles, one minute as 1.15 miles, and one second as roughly 101 feet. For a map, a second matters. For a photograph of a table leg it is far below anything the pixels can support, a point worth returning to.

There is one practical hazard in the notation. Rounding each part separately can produce 59.997 seconds shown as 60.00, a value that should not exist. The honest way is to round the whole angle once to the smallest step and then cut it into parts, so that 12 degrees 34 minutes 59.997 seconds becomes 12 degrees 35 minutes exactly.

What the camera does to an angle

Now the trap. A camera flattens a scene onto a plane. If the angle being measured lies in a plane that faces the lens squarely, the flattening changes its size and nothing else, and size does not matter. If that plane is tilted away, the picture is squeezed in one direction: across the axis of the tilt, by the cosine of the tilt. A squeezed angle is a different angle.

How different depends on how the angle sits relative to that axis. A right angle with one side lying along the axis of tilt still photographs as a right angle at any tilt. Turn the same corner so the axis halves it, tilt the plane by 60 degrees, and it photographs as 53.13 degrees. Turn it a quarter turn further and it reads 126.87. Same corner, same tilt, three answers.

Sports science has run into this repeatedly. A study in PLOS ONE filmed a fixed figure from four perspectives and compared angles taken from the video in the Kinovea program with a drawing reference. The authors found the software valid and reliable at distances up to 5 meters and across camera angles from 90 to 45 degrees, and still recommended the orthogonal view, 90 degrees, for the best results.

The human body makes it harder, because limbs rotate out of the plane on their own. Researchers comparing two-dimensional video with three-dimensional motion capture during drop jumps reported that the frontal plane projection angle of the knee, a popular measure taken from a single camera, showed poor to moderate agreement with the three-dimensional values, with intraclass correlations from 0 to 0.57. Their explanation is plain: a flat image cannot tell rotation of the leg about its own length from sideways collapse.

None of this makes photographs useless. In a hand clinic, 50 patients with contracted fingers were photographed with phones and measured on screen; across 123 joints the mean difference from a manual goniometer was 3.2 degrees, inside the 5 degrees usually accepted as the error of the finger goniometer itself. The patients had been given written instructions and sample pictures showing how to hold the hand. The method worked because the pose was controlled.

So the rule of practice is short. Put the camera square to the plane of the angle, as far back as is practical, with the corner near the middle of the frame. If that cannot be done, treat the number as a description of the image.

Carpentry, where the plane is yours to choose

Workshops have an advantage that clinics lack: the object holds still and can be laid flat. A miter is the simplest case. Two boards meeting in a square corner are each cut at half of 90 degrees, 45. A six-sided frame turns through 60 degrees at each corner, so each cut is 30. Photograph the finished joint from directly above and the picture can check the saw.

Roofs are described as a ratio instead. The United States workplace safety rules define a low-slope roof as one with a slope less than or equal to 4 in 12, vertical to horizontal, and a steep roof as anything greater. As an angle, 4 in 12 is the arctangent of one third, 18.43 degrees. A gable photographed square-on from across the street gives that angle directly; photographed from the corner of the lot, it does not.

A worked example, and what a pixel is worth

The sample picture in the tool on this page is a triangle with sides 3, 4 and 5, drawn 800 pixels along its base. With the vertex on its sharpest corner and a point on each side, the reading is 36.87 degrees, which is the arctangent of three quarters. The same angle is 36 degrees 52 minutes 12 seconds, 0.6435 radians, or 40.97 gon. Its supplement is 143.13 degrees and its complement is 53.13, the other sharp corner of the triangle.

The three-point method is the standard one; the ImageJ program long used in laboratories describes its angle tool as measuring an angle defined by three points. What differs between tools is how well the points can be placed. Enlarged images are normally smoothed, and smoothing blurs the very edge being aimed at. Browser documentation notes that turning smoothing off keeps the pixels sharp, which is what a magnifier for placing points needs.

Then comes the arithmetic of a pixel. Move the far end of an 800-pixel side sideways by one pixel and the angle shifts by 0.07 degrees. Do the same on a 100-pixel side and it shifts by 0.57. Long sides are the cheapest accuracy available: mark points far from the corner.

Finally, tilt. The tool’s three-dimensional gauge shows the same 36.87 degrees on a protractor base. Seen face on, it reads true. Slide the camera tilt to 60 degrees, from the side the view opens on, and the angle on screen is 28.72. Nothing about the gauge has changed.

That is the limit of the method, stated without decoration. One picture cannot recover the true angle of a tilted plane, because the tilt is not recorded in it. Lens distortion near the edges of a wide shot is not corrected either. A measurement taken from a photograph is a careful statement about a photograph. Making it a statement about the world is the photographer’s job.