here is the final version of my vehicle speed estimation demo read the thread below to learn how I built it. I will cover: - detection - tracking - perspective transformation - speed calculation - some bonus ideas ↓

1. before diving into speed estimation, we need to solve the detection of moving vehicles. I used YOLO-NAS. it's capable of performing real-time detection on low-compute devices, including the Jetson Nano. link:

2. for object tracking, I used ByteTrack. this enables us to assign a unique ID to each vehicle. doing so, we can accurately track the distance traveled by each car, forming the basis for subsequent speed calculations. link:

3. calculating the distance traveled can be challenging due to the camera's perspective. car moving at a constant speed will appear to move a different number of pixels in the image, depending on its distance from the camera.

4. we can precisely determine the position of each vehicle within our zone. now, each vehicle is represented by x and y coordinates, expressed in meters. this allows us to compare a vehicle's current position with its position from a second ago, enabling the speed calculation.

5. knowing the exact position of each vehicle opens the door to more advanced applications. for instance, we can now accurately estimate the distance between vehicles. we can see that the driver of the white car came dangerously close to the vehicle in front of him.

6. we can also take this a step further and visualize the entire road in 3D, showcasing the vehicles as they move along it.

Somehow reminds me @comma_ai calibration challenge

@comma_ai Nah… what they did is a lot more complicated :/

this is wild

thanks a lot! noting compared to LMMs magic you guys build