https://www.youtube.com/watch?v=ZzMdVqVliqs
We already keep track of the most big near-Earth asteroids. But when it comes to small sized ones – we are basically ignorant of them. For example, we estimate that there is at least 1 million of 40 m near Earth asteroids, and we discovered at most 1% of them.
In order to detect an asteroid we need a telescope. All ongoing asteroid detection projects(NEAT, LONEOS, CINEOS, etc.) use big and very expensive telescopes. But even with big telescopes it is hard to detect 40 meter rocks tens of millions kilometres away.
We propose a change of paradigm – by using thousands of low cost telescope we will watch out for asteroid occultations. Star shadow of asteroids moving across Earth surface will create a pattern of occultation detections in our telescopes array and by using machine learning techniques we will produce current asteroid celestial coordinates.
The good news is that we can get a lot of telescopes for free – there are thousands of amateur owned computer controlled, digital camera compatible telescopes! All we need to do is to help amateurs to connect them to Internet!
The following software implements detection of asteroid shadow despite noise in telescope CCD.
- python
- Java 7
Simulation is implemented as python script.
Solution is a Java program.
Compile Java code and execute:
java solution.SmartSolutionManager NUMBER_OF_ITERATIONS
NUMBER_OF_ITERATIONS is an optional parameter indicating the number of times the algorithm will be executed.
That will run simulation and solution. A new set of data input files is generated on every iteration.
The output of the program is result.txt file containing the list of errors for each iteration and also the number of iterations with no solutions.
As an input we have some 10 000 detected events (occultations) per second. Almost all of them are noise. We simulate passing of asteroid with our software and put generated events on our processing algorithm. Then we calculate error by comparing simulated asteroid and extracted asteroid data. As test runs show, 50% of time we recover almost perfect data, and 40% of time error is below 10% which proves that solution is acceptable.
We go through all pairs of events detected on edge of the array. Taking only events on the edge and filtering by time reduces search space to some thousands of possible pairs.
Then we estimate possibility that this pair of events was generated by asteroid by going through events in-between. If we detect some events in-between then we perform extra processing by gradient descent and output solution.
See sample input and output data within the doc folder.