Why the effects of data delay and loss are not usually considered in the design of multi sensor trackers but the robustness of the designed trackers is investigated against data delay and loss in the simulation?
For example, "Frame Loss" is not considered in the design of vision based trackers but in the simulation phase is considered.
Dear Ramin,
the design phase tends to idealize the input. Thus input errors are not considered. ("Better" designs take input errors to be expected with some probability into account!)
When it comes to "evaluation", errors occurring with some likelihood are introduced to assess "overall performance".
From my own experience: tracking algorithms may be robust against frame losses though not specially designed for. It depends ...
Hello. The fact that you are referring to delay and loss means that you are assuming that generally, these factors are not a problem, but they sometimes occur. You could ask the same question about the communications backbone that underlies the signal processing: why doesn't it allow for lost data, etc?
You in fact can deal with such problems by introducing redundancy into the system: e.g., multiple cameras, data buffers and by processing with a delay equal to the maximum delay that you expect to occur. Obviously this has drawbacks in terms of how "up to date" the tracker output is and how complex the system and processing have to be. You are free to design a tracker according to whatever assumptions you consider realistic.
By the way, when you talk of the "design" of multisensor trackers, that is maybe a bit generous since there aren't really any design methodologies. It's usually simulate and see. It is difficult to analytically predict Kalman filter performance under data loss situations, much less the performance of a multisensor tracker, without performing Monte Carlo simulations of it. Hence it's still in the realms of a black art to some extent. You can get lower bounds on errors, but these are no use for predicting actual performance.
Have a look at the reference below, you will understand what I mean. Not a single one of these methods has a a design methodology associated with it!
The other link has some information about robustness against faults in a navigation system.
Article Taxonomy of multiple target tracking methods
Conference Paper Optimising the algorithm design for high-integrity relative ...
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Computer Graphics and Animation
- 3D