Hamamatsu UVtron Flame detector
The Hamamatsu UVTron flame detector (available from www.acroname.com) is commonly used in the Trinity Firefighting contest to detect the presence of a candle. It is most commonly used just to determine whether a candle exists in the room being inspected; less often used to determine the direction of the candle.
It senses ultraviolet light which is emitted in the candles flame. Ultraviolet is apparently not emitted from other sources which may be expected at the contest. Including, I assume, camera range detectors (which use IR) and the sodium vapor lights which are used to light the gymnasium at Trinity. I am using it successfully in a room which has plenty of indirect sunlight, occasionally direct sunlight or incandescent light. (Flash photography is another light source which is an issue at Trinity. I don’t know if it emits UV or not.)
Characteristics of the UVTron are that it can detect a candle quite reliably at the distances necessary in the contest (up to 4 or 5 feet). It is even able to detect a candle by reflected light off the room walls when the candle is not in direct view. My own experience is that it can always detect a candle in direct view. It seems able to detect a candle where direct view is blocked by a piece of "furniture"; in this case, the reflection is off walls that are close to the candle and probably result in a good reflection. (the candle is usually in a corner, with the furniture nearby between the candle and the doorway). If the candle light must travel to a far wall (3 feet away) and then reflect back to the sensor, the probability of getting a detection becomes less.
The UVTron provides an output consisting of 10 msec pulses (either from ground to +5, or +5 to ground, or open collector…you pick). The rate of these pulses indicates the strength of the signal. However, rate isn’t as useful as it might be as the pulses will be output at a minimum interval of 160 msec. And since you don’t want to spend too much time waiting to see if there is a detection , you can’t expect more than 6 pulses in a one second inspection.
A weak candle signal may take a whole second to show up. And, by the way, the UVTron apparently has some logic built in such that if it doesn't get 3 detections over 2 seconds, the board will ignore the detections. My interpretation of this is that if you get one pulse out of the sensor, there have been 3 detections in the last 2 seconds…which sounds like a good detection level.
The UVTron is not commonly used for candle direction. I believe this is due to two factors: one is the slow output pulse rate which would make scanning the room a very slow operation; and second, the ability to pick up reflections might cause uncertainty as to whether the candle or a reflection is being detected. A sensor with signal strength output is better for sensing direction. (e.g. the Eltek pyrosensor described elsewhere)
The UVTron is most sensitive over about a 90 degree arc straight ahead (+/- 45 degrees). Hence, aiming it at one direction in the room may not detect a candle located 90 degrees to the side. I dealt with this by having the robot enter the room and do a 1 second 180 degree pivot to point out of the door again. This allowed enough time for a reliable detection, pointed the UVTron all over the room during the pivot, and left the robot aimed to depart if a candle was not detected. During the pivot, the robot also scanned the room with the Eltek sensor; and if the UVtron reported a flame in the room, the robot would quickly orient to the direction in which the Eltek had detected the strongest heat source.