speed-camera
speed-camera copied to clipboard
Calibration with different Screen Ratio and Framerate
Hi
I used this tool currently in a docker testing envoirnment with version 11.22 (see #155 ).
My camera ist almost 90 degree to the street and in around 10m distance. I use a Reolink 810 with RTSP, the size ist 640x360 which not fit to any of the plugins. The framerate ist 15 fps
Calibration ..... cal_obj_px_L2R=160 px cal_obj_mm_L2R=4700 mm speed_conv_L2R=0.10575 cal_obj_px_R2L=150 px cal_obj_mm_R2L=4700 mm speed_conv_R2L=0.11280 (Change Settings in /root/speed-camera/config.py) Speed Trigger ... Log only if max_speed_over > 10 kph and track_counter >= 6 consecutive motion events Exclude Events .. If x_diff_min < 5 or x_diff_max > 40 px If y_upper < 35 or y_lower > 160 px or x_left < 200 or x_right > 520 px If max_speed_over < 10 kph If event_timeout > 0.50 seconds Start New Track track_timeout=0.30 sec wait after Track Ends (avoid retrack of same object) Speed Photo ..... Size=640x360 px image_bigger=1.0 rotation=0 VFlip=True HFlip=True image_path=media/images image_Prefix=speed- image_font_size=14 px high image_text_bottom=True image_jpeg_quality=95 image_jpeg_optimize=True Motion Settings . Size=640x368 px px_to_kph_L2R=0.105750 px_to_kph_R2L=0.112800 speed_units=kph CAM_LOCATION= None
As you see I set the motion track area to a 320x125px area. Cars are usually within 140-170px - I set 160px I don't touch the cal_obj_mm_L2R / cal_obj_mm_L2R yet I I think this should work out.
The results I get are > 150 km/h up to 600-700 km/h if it save something or it simply I dont get any result. Some delivery guy was tracked with >80km/h walking speed ...
Here some Log maybe I overseen something. speedcam_log_20231128.log
Here on the street are 30 km/h allowed and I guess it should not be over 50 km/h but by this huge difference I wonder where I have done the mistake in the settings
Thanks :)
So after some days with irregular results, I found my error. I don't get that I should calibrate with the detected size (green frame) and not with the real car size. As I have quite different size of the detected area/part It was hard to say the size correct.
So I used the area I want to use for detection 220px with are 4900mm - as R2L and L2R is the same distance (small street almost one lane) I think this would not make much differences.
I get now maybe 40-50% correct results or at least results that I would think could be correct. But still a lot of cars drive here above 300km/h :)