Demo of vehicle tracking and speed estimation at the 2nd AI City Challenge Workshop in CVPR 2018
HTML-код
- Опубликовано: 14 окт 2024
- In this demo video, the estimated speed of each vehicle is shown in miles/hour.
Our team from the University of Washington is the winner of Track 1 (Traffic Flow Analysis) at the 2nd AI City Challenge Workshop in CVPR 2018.
Our source code is available at the following link:
github.com/zhe...
The datasets for the 2nd AI City Challenge (2018) are no longer available to the public. However, you may still access the datasets of the 3rd AI City Challenge (2019), or participate in the latest 4th AI City Challenge in CVPR 2020. They provided a new city-scale dataset for multi-camera vehicle tracking as well as image-based re-identification. They also had a new dataset for traffic anomaly detection. The scale of the datasets and the number of vehicles that are being used for evaluation are both unprecedented.
The website of the AI City Challenges is at www.aicitychal.... Please forward your inquiry to aicitychallenges@gmail.com.
good work Zheng Tang. quite an inspiration
Awesome work @Zheng Tang & co. - Very impressive.
Am I correct in thinking this software could run a count on any piece of recorded video? And do you have the ability to export a final breakdown of the the various counts? Our local BIA is debating installing a bike lane / we're currently just using the pneumatic hoses for counts, would love something a little more reliable / detailed like this -
Cool video. Nice colors! Is there some reason each car has a different color? Thanks!
The colors are just to indicate different vehicle identities.
@@DrTangerineAI Interesting! It is good way to separate them. We use colors of trajectories to indicate the category of the vehicle detected.
How can I find a ready-made template so I can count the number of people for my project? in the Roboflow???
How can you compensate the image distortion due to the perspective view? Simply finding the change of pixel could not actual calculating the vehicle speed.
I have worked a project about intersection collision prediction system, a method called "Distance map" is generated with big data and DL, to add a weighting to each pixel on the captured view,
this project won an award from Geneva inventions Bronze Medal.
Congratulations on the medal! For camera undistortion, I simply applied cv.undistort(): docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html
@@DrTangerineAI ruclips.net/video/mHkz8yT_vxY/видео.html
Thanks. This video shows the concept what I have applied to my project for actually detect the vehicle speed
Nice video and nice project . Do you have source code to detect Number plates?
Hi This is great work !
One question - How are you comparing your predicted speeds to the actual speeds ?
Is there a dataset for that or are you just predicting speeds with your algo since the data is not available ?
@@SouvikBagchi There is no public benchmark with fixed camera view available now. However, you can experiment with autonomous driving datasets that have moving camera views.
Apa nama dan Spec alat ini, saya mau beli, terimakasih
So smooth... Can you tell me FPS ?
We demonstrated on a Jetson device in nearly real time (10 FPS).
Hi @Zheng Tang thank you for making this project opensource , i have a question aboutrun , how can i put the input video? , and its possible to allow camera ip With RSTP protocol ? thank in advance
You can follow the guideline in our GitHub repo: github.com/zhengthomastang/2018AICity_TeamUW
@Zheng Tang, Could you please let me know about the data collection? What camera was used? Also could you please suggest a camera for night-time data collection ?
Did you get the answer or any solution to the problem of collecting dataset?
@@talhayousuf4599 I need on-road data and not from a height, so I used an ordinary dash cam fitted on car's dashboard.
how to determine the position of each object identified in the video in the output like this
bus: 89% , Position : (0.032253, 0.110209), Height and Width : 0.063911, 0.072384
Your work is really amazing. Can i know on what hardware did u run this.please let me know.
Zheng Tang appreciate for your reply
What you used for recording video
Hi...
Thank you for the tutorial.. It's impressive. I have one question.
Can we make the animation of moving objects a function of time? the program will run and continuous animation will run???
Thank you
You can find our GitHub repo here: github.com/zhengthomastang/2018AICity_TeamUW. The input is the original video and the output includes the coordinates of bounding boxes at each time index. The animation can be produced online or offline, which is just for visualization purpose.
where is the tutorial ?
He is very greedy, just showing video is not a great fun.
May be this video belongs to someone else.
I disliked this video completely. I am sorry.
@@bolzanoitaly8360 wtf bro do you want a massage too while going ?
Nice implementation.. Do you have similar source with Python and Tensorflow? like resnet or with efficient net
Sorry but we only have the C++ version available.
@@DrTangerineAI can you please send me the C++ version i didn't find it anywhere thanks !
@@bendahouwael1176 github.com/zhengthomastang/2018AICity_TeamUW
Pls try in ANN
How do I get a copy of your program, Demo of vehicle tracking and speed estimation at the 2nd AI City Challenge Workshop in CVPR 2018
Hello Zheng, Can you please help me to with steps with how I can run your code to see the output of track 1. I have seen six packages in the code but not understanding the Main file to be executed. Can you please write one "readme" file for steps of execution. It would be very useful. You have done a great work !!
Thanks a lot for quick reply..I will try it and let you know
Hi Zheng Tang, thank you for sharing the code . I am using faster cnn and i do not know how to give an id to object do you have any opinion? Best
@@DrTangerineAI thank you for your reply. I looked your github repository but I don't understand clearly .Can you briefly explain it ?
Hii I know about this project.ill do image annotation, lanes and bounding boxes, how to do this project on online.
Hi! kindly let me know of the frame rate your algorithm is capable of running at. Can it handle occlusions?
Can it be used to handle multiple cars.
I would like to know if you could share the original video where to perform the test?
I already saw the note in your repository, thank you very much for the note.
Hi Sir ,I see your publication at CVPR 2018 , do you integrat ALPR in your source code?
Accoding to your description, we are to our inquiries to aicity but i hope you won't mind answering this question: I'm working on a project and I would like to track vehicles. If I contact aicity and ask them for access, will they give it to me or do i have to participate in the 2021 challenge first? Don't mean to repeat the question but you said we can still access 2019 datasets. Where? Can you please provide the link?
As for your source code, thank you for sharing. Is there a document I can sign for you if and when I use your code? In your GitHub link, you asked us to site your paper. Is there something more official I can do for you? And would it be possible to contact you here again for inquiries about your code?
Thank you again.
Our demonstrated experiments were conducted on the 2018 version of the AI City Challenge data sets. From what I learned, NVIDIA cannot release the 2018 version any more. However, the 2021 version is currently available, which is an expanded set of the 2020 version and 2019 version. They now have a much larger city-scale data set along with better annotations. Hope it will still help your research.
Please email to Prof. Jenq-Neng Hwang (hwang@uw.edu) if you want to use the code for commercial purpose. For paper citation, you may ignore it if you are not publishing a conference/journal paper. For other technical questions, please add them to the Issues of our GitHub repo, so that other people can see our answers as well. Thanks.
@@DrTangerineAI Thank you for the prompt response. I understand that 2018 is no longer available but I wanted to know if all i had to do to get access to those datasets is to literally request it from them by email instead of having to participate ie datasets are only available to participants.
I'll look at your code and post in the ISSUES section just like you asked if I run into anything. Thank you again for responding and for the email.
Hey, great work. was this done on real time or prerecorded video? Can we implement in python too? what all things are needed?
Here is the GitHub repo: github.com/zhengthomastang/2018AICity_TeamUW. It is an offline method. Since the code is broken down into multiple parts, it is hard for us to measure the runtime.
can I get the paper or source code for the winner of this challenge?
That is wonderful. I want to read your paper. Can i read it?
Hello man, I m looking for this projects, I need to count vehicles, motorcycles and persons for my projetcs job. Tell me if can We talk
you have this video sample?
Great job Thomas!
ага александар
is the vehicle and speed detection in real time. what hardware was used to accelerate the process
Our method is offline. We used NVIDIA GPU for object detection.
Hello, Dr. Tang can I ask you if you used a GPU for the training model?
If yes, which one Thank you.
If I remember correctly, we used a 1080 Ti for training at that time. There have been a lot of new versions now.
please someone should tell me how to get those flying lines....
what is your accurancyin total?
Could more information be given how the speed is calculated ?
Zheng Tang thanks
@@DrTangerineAI Could i possible get the the dataset from you ? many thanks
whats is the algorithm to estimate speed...please help me sir?
You can check our GitHub repo here: github.com/zhengthomastang/2018AICity_TeamUW
I hope in the future you can make a tutorial explaining the code. it would be fantastic
Please check the repo at github.com/zhengthomastang/2018AICity_TeamUW.
Me need this software
any paper on this program i can refer? i want to learn how this works
openaccess.thecvf.com/content_cvpr_2018_workshops/w3/html/Tang_Single-Camera_and_Inter-Camera_CVPR_2018_paper.html
@zheng do you have the some code developed with python
@@DrTangerineAI I'm developer python , i have big problems C++ : when i run this command g++ main.cpp i find this issues :
/usr/bin/ld: /tmp/ccoPmRqH.o: in function `cv::String::String(char const*)':
main.cpp:(.text._ZN2cv6StringC2EPKc[_ZN2cv6StringC5EPKc]+0x4d): undefined reference to `cv::String::allocate(unsigned long)'
/usr/bin/ld: /tmp/ccoPmRqH.o: in function `cv::String::~String()':
main.cpp:(.text._ZN2cv6StringD2Ev[_ZN2cv6StringD5Ev]+0x14): undefined reference to `cv::String::deallocate()'
/usr/bin/ld: /tmp/ccoPmRqH.o: in function `cv::Mat::~Mat()':
main.cpp:(.text._ZN2cv3MatD2Ev[_ZN2cv3MatD5Ev]+0x39): undefined reference to `cv::fastFree(void*)'
/usr/bin/ld: /tmp/ccoPmRqH.o: in function `cv::Mat::create(int, int, int)':
main.cpp:(.text._ZN2cv3Mat6createEiii[_ZN2cv3Mat6createEiii]+0x9d): undefined reference to `cv::Mat::create(int, int const*, int)'
/usr/bin/ld: /tmp/ccoPmRqH.o: in function `cv::Mat::release()':
main.cpp:(.text._ZN2cv3Mat7releaseEv[_ZN2cv3Mat7releaseEv]+0x4b): undefined reference to `cv::Mat::deallocate()'
/usr/bin/ld: /tmp/ccoPmRqH.o: in function `cv::Mat::operator=(cv::Mat&&)':
main.cpp:(.text._ZN2cv3MataSEOS0_[_ZN2cv3MataSEOS0_]+0xe7): undefined reference to `cv::fastFree(void*)'
collect2: error: ld returned 1 exit status
please help me
Hi @ZhengTang , tell me which lines in your code , you put ""the line direction in the rear of vehicle "" thanks
Hello @Zheng i mean after bounding box of vehicle how can you draw directions of vehicle with colors .
@@DrTangerineAI please could you please take a look in this project , i want to draw directions of vehicle with colors like your project , github.com/JdeRobot/traffic-monitor
@@DrTangerineAI please i need your help . jut 5 min of your time
@@DrTangerineAI please check issues
Sir please see this code i tried since 2 weeks to draw directions of vehicle with colors with function cv::line , unfortunately i haven't succeed , could you please give me 5 min .
You used a high spec desktop in those experiments, I was wondering if the track 1 experiments could be done on a mid-low spec machine, I want to try it since I work in a traficc monitoring enterprise and have access to street videos.
@@DrTangerineAI Thomas thank u for the answer but I have another doubt, do I need Cuda for this project? because I have an AMD GPU so no cuda for me.
this code is not working. how to make it worked
insane, dude
Very impressive!
do you have the code version python ??
Sorry. Our code was developed in C++: github.com/zhengthomastang/2018AICity_TeamUW
HI Zheng Tang
when i run the folder of speed_detection it give me : Error: camera parameters not loaded ,
@@DrTangerineAI i try to generate parametre of camera , its give me this message : ""Reprojection error is small enough. Stop generation"". please i need your help
@@DrTangerineAI thank you so much
@@DrTangerineAI please can you do a video for explain that , it does not work for me
Sorry for the late response. You have to execute the code in order, since the output of a module can be the input of another module.
@@zakariaharakat7759 This is the expected message as the objective error is small enough. The generation stops, and you can collect the output results.
Thanks 🙏
have you checked your result?
Yes. You can find the results in our paper: openaccess.thecvf.com/content_cvpr_2018_workshops/w3/html/Tang_Single-Camera_and_Inter-Camera_CVPR_2018_paper.html
How about source code c# with yolo?
We used YOLO for detection of objects in each frame. But the tracking part was implemented separately.
Is it real time ?
Hi, can you share me the code of MATLAB. I need this code for my final year project. Thanks
Shoutout to the chad going 69.420 on the highway
how can i do the same job with yolov5 , please help
Please check the repo at github.com/zhengthomastang/2018AICity_TeamUW. You may change the YOLOv3 component with YOLOv5.
@@DrTangerineAI ،i am sorry, but the code is written in c, i don't know c language, is there away that i can find one written with python
@@elbaz_afandy I may release a Python version in the future. Will keep you updated.
@@DrTangerineAI thank you very much,
Try speeding and getting away with it now LMAO...
These are not approximate just estimates. You could beat that in court. I think this project is to determine the flow of traffic.
and now Big-brother is really watching you hahaha.
LOL
hahah
Amazing!