Hey there! In this repo, you'll find a tool that our group put together to transform video data we captured with two drones into CSV-based car-following trajectories. We got our footage from Park St in Madison, Wisconsin. We're using this transformed trajectory data to test out a bi-scale calibration approach for car-following models. We've been working on a paper that describes the detailed methods behind this tool, and we're planning to upload it to arXiv soon. Once our paper gets the green light, we'll be sharing the video data here too. Stay tuned!
The studied area of the data collected from Park St in Madison, Wisconsin
Here are the prerequisites for using this repo:
- Python 3: Ensure you have a Python 3 environment set up.
- Required Packages: Install all necessary packages listed in the
requirements.txtfile. - YOLOv7: If you're keen on executing the detection component, YOLOv7 needs to be installed.
To use this repo, run each Python script in the order we introduced in this section. As you proceed through each Python script, always verify the paths for both the input and output files. This ensures that everything runs smoothly.
The process of vehicle trajectory extraction and cleaning contains two parts: trajectory extraction and trajectory processing. Please find the details of each steps in our paper.
All code in this steps is under folder "TrjExtraction".
Step (1): Selects keyframes from the vedio. (./TrjExtraction/1 ExtractFrame.py)
Step (2): Stabilizes all frames by matching the feature points. (./TrjExtraction/2 PicStable.py) [💡Notice: If your camera is fixed and the frames are stable, please skip this step.]
Step (3): Merges the frames from different drones and gets the full scope of the target range. (./TrjExtraction/3 Merge.py) [💡Notice: If you only have a single video source, please skip this step.]
Step (4): Detects and tracks the vehicle using locally trained YOLOv8, then get the initial trajectory data of all vehicles in pixel coordinate. The code for this step is not listed. Please refer to the YOLOv8 tutorial. [💡Notice: We highly recommend using YOLOv8 or later versions, as the tracking functionality is integrated into the model starting from YOLOv8, eliminating the need to set it up separately.]
All code in this steps is under folder "TrjProcessing".
Step (0): Converts the trajectories to csv format. (./TrjProcessing/0 trackresult2csv.py)
Sept (1): Determine the edges of each lanes in the map. (./TrjProcessing/1 Determine edge and lane.py)
Step (2): Transfers the coordinate trajectory to Universal Transverse Mercator Grid System (UTM) coordinate trajectory. (./TrjProcessing/2 Pixel2Global xy.py)
Step (3): Removes position offsets and then smoothed the trajectory. (./TrjProcessing/3 Trj Clean.py)
Step (4): Calculates the vehicle speed, acceleration, and position. (./TrjProcessing/4 Determine va.py, ./TrjProcessing/5 Veh_info.py, ./TrjProcessing/6 Determine Preceding veh.py)
We will provide the processed data soon. Data attributes are shown below.
| Attribute | Unit | Description |
|---|---|---|
| id | - | Id of vehicles |
| time | - | Time in the format of YYYYMMDDHHMM.S. |
| x_pix | pixel | The horizontal pixel coordinate of vehicle |
| y_pix | pixel | The vertical pixel coordinate of vehicle |
| w_pix | pixel | The width of an object in pixels. |
| h_pix | pixel | The height of an object in pixels. |
| edge | - | The position of edge or intersection. |
| lane | - | The position of lane at given edge. If the vehicle is in an intersection, lane is 0. |
| x_utm | m | The UTM x-coordinate of vehicle |
| y_utm | m | The UTM y-coordinate of vehicle |
| t_sec | sec | The record time in seconds |
| v | m/s | Speed of the vehicle |
| a | Acceleration of the vehicle | |
| pre_id | - | Vehicle id of the previous vehicle |
| pre_v | m/s | Speed of the previous vehicle |
| delta_d | m | Distance between the outer contours of the subject vehicle and the preceding vehicle |
output_video_s.mp4
Developer: Keke Long (klong23@wisc.edu).
Code reviewer: Hang Zhou (hzhou364@wisc.edu).
If you have any questions, please feel free to contact CATS Lab in UW-Madison. We're here to help!