It goes up/down in a straight channel? Use Aaron? ADX? I don't think a computer program can recognize trends, at least not until the trend has gone a long way. If most traders can't recognize the earlier stages of a trend, just by using TA; how could you expect a machine do so? You can only forecast the earlier stages of a trend by using fundamental analysis.
If a computer program cannot recognise it, then it is not there! What does a human use to detect the trend? price/time data. That is exactly what could be programmed into a computer.
Use the P, V relationship as the gross objective measure. Trends overlap for a period of time. The period before BO and after the over lap begins can be handled by using a less gross measure but you do already have a 100% definition by just using the gross measure. The P, V relation is not complete and for the case where it has a void the situation is trivial so you are still covered. The relationship was used before it was published in 1934, so it has been in general use for quite a while and through most market behavior including Black Swan classified type stuff. If you use this stuff it is off scale vis a vis your PF poll. That is probably why you are observing the results you are getting so far. There is a very large shift in performance between edge trading and trend trading. Edge trading is at the low end. If you look at edge oriented scripts by a person like Atchuler (Wealthlab), you can easily see why. Also read Ross to discover his limitations as to understanding when trends are in effect. He rarely sees them. There is always a trend in the market as shown by the P, V relation. The best appraoch is to apply the relation to each of the total range of practical fractals (seven).
You need to, with a pen and paper, decide what characteristics your trend has. Eg todays 200 day moving average is higher than yesterdays 200 day moving average. The 100 day moving average is 15% higher than the 200 day moving average. etc I have coded a custom indicator that uses moving averages, volitility, standard error, time and volume. That works well for me, but I'm not in the US. For starters: Google 'Guppy Multiple Moving Average'. It will give you an example of what you might want to achieve. Good luck, pneuma
In the mode of the thread. Keep using the pen and paper. Write out the P, V relationship so you can code it. to make it easy, use two "if, then" Boolean statements. then you can "or" them together and use the output of the OR to tell you when there is a trend, when it ends and wnen the next trend begins in earnest. Trends overlap it turns out and that makes it a fortunate situation for you because there is always a trend being expressed.