Usually you'll have too few trackers to Solve if you increase the track filtering to a point where too many trackers are removed / ignored by the solve.
3D Camera Tracking is a process of fine tuning everything from the tracking parameters, the camera focal length (if known), and solve filtering.
It's really important to try and remove areas from the track where the movement is non-camera, and in this case you should mask out any moving traffic and people walking along the street.
If the CameraTracker locks onto those, then your solve will never be any good.
To create occlusion masks for the CameraTracker, draw and animate a polygon over the area you want to ignore and INVERT it before piping in to the Track.Mask input.
If you connect more than one polygon to mask out unwanted objects, the top poly must be set to Invert, then any others following must have their paint mode set to subtract.
I found and downloaded the footage you used from YouTube, and have solved a track for you, using the frame range in your comp. Solve result is 0.36 which is pretty good.
You can download the setup here;
https://drive.google.com/file/d/1IubHBt ... sp=sharing
Windows 11 // i9-13900K // RTX 4080 // 128GB DDR5 RAM // 2TB WD M.2 NVMe // ASRock Z790 // TH420 AIO Liquid CPU Cooler
Mac Ultra Studio M1 64GB