Hello, I am working on car detection. My goal is detecting the cars which has top view. Because the size of images are high (3500*4000 pixels), using RCNN is not possible (it takes more time for training). Now, I am using ACF for first detecting. I have some problems: 1- For post-processing which method is suitable(I use HoG + SVM but it does not good response)? 2- Is there other ways for this work?
Thanks.
Dear Shiva Mlk ,
I didn't understand your problem. Are you looking for alternatives to ACF detector? Did you try to train an ACF as top view car images?
I think you can obtain better results by changing some ACF's parameters, for instance, the cascade threshold and the number of trees. There are many methods (handcrafted-based) which achieve better results than ACF (see http://www.vision.caltech.edu/Image_Datasets/CaltechPedestrians/rocs/InriaTestRocReasonable.pdf), however, the codes are not publicly available.
Finally, I recommend you take a look at this paper: "https://rodrigob.github.io/documents/2014_eccvw_ten_years_of_pedestrian_detection_with_supplementary_material.pdf"
It provides an excellent review of methods for sliding window detection.
I hope this helps you
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training