I have small problem that requires to process both spatial and temporal information.

I need to predict vehicle's trajectory based on previous trajectory information and map information. My current situation:

  1. Predicting future trajectory of vehicle was successfully done using seq2seq model. Where I give previous N points and model returns K possible trajectories
  2. I have tried crop map information(based on vehicle's position) and giving it to CNNLSTM model with trajectory information. But even test accuracy was not good.

What architecture is best for my problem?

P.S I am new to this forum. Feel free to improve question or point out my mistakes.

  • $\begingroup$ What are your input and output shapes? $\endgroup$ – SaTa Oct 24 '19 at 3:59
  • $\begingroup$ @SaTa My model is not complete. Question is about model design. I have sequence of coordinates and map image. Output is sequence of coordinates. $\endgroup$ – Elbek Oct 24 '19 at 4:38
  • $\begingroup$ I know. The architecture depends though on what the inputs are and what output you are trying to predict. What do you mean by a map image? Do you have an image per each time step? $\endgroup$ – SaTa Oct 24 '19 at 4:41
  • $\begingroup$ I have N sequences of vehicle coordinate and map image for each record. I want to predict next N coordinates of vehicle $\endgroup$ – Elbek Oct 24 '19 at 5:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.