tensorflow_object_counting_api
tensorflow_object_counting_api copied to clipboard
ahmetozlu
Inaccurate count
When I run vehicle_counting. py,so I would like to ask where in the code the range of counts changes and I found that the pedestrian count was accurate and the car count was not. In addition, I would like to ask what is the benchmark for counting cars through the line, and how can I achieve count=1? I feel that there is something wrong with the benchmark of this range
The vehicle_counting.py fixed, can you try again to run?
"In addition, I would like to ask what is the benchmark for counting cars through the line, and how can I achieve count=1? I feel that there is something wrong with the benchmark of this range" -> What do you mean? Can you specify it more, please?
@ahmetozlu I found the output video: the count is 3 when a car passes, which is not accurate, so I would like to ask how to modify the counting area,and hit the second car past time to become 5 I noticed that there was a purple box marked "surfboard" when the car passed. I don't know what it means. Is it the designated area of interest when the car passes
https://github.com/ahmetozlu/tensorflow_object_counting_api/issues/44#issuecomment-521876648
@ahmetozlu Thank you. I just tried to change deviation and it has some effect. I will change the model later In addition, I would like to ask how the count is counted. I did not understand the code. My understanding is that when the car passes, the ROI line turns green and count plus one. Because the next frame may also be when a car passes, and these two adjacent frames are when a car passes, how do you determine that it's the same car
This is a great question, thanks: "Because the next frame may also be when a car passes, and these two adjacent frames are when a car passes, how do you determine that it's the same car"
My answer: The current version tensorflow_object_counting_api can not understand that is same car/object or a different car/object because it has not a tracker system yet. However, I almost finalize the "multiple object tracker module", which is a Kalman Filter based multiple object tracker implementation, and after I integrate it to the tensorflow_object_counting_api then API will be able identify/track multiple objects and it will not count the same object more than 1.
@ahmetozlu Thank you for your reply. I would like to know how to mark the car as completely passed or start to pass the kalman filter principle marked as 1 when it passes the ROI line. Do you have any related papers for this project? Because the code I do not quite understand why you have set (top+bottom)/ 2-roi
me too，If someone stands nearby the roi line, there will be a lot of count, and this problem will be even more serious if it is counted through webcam or ipcamera
@cranehovers yeah,Have you figured out how to solve the problem
I think we can solve this problem through multiple object tracking (MOT)，and i'm trying it