Advice on tuning for cars

[bushibot]

I wonder if there might be a better place to ask? but I'm having some tuning configuration challenges. I need to catch vehicles moving L-R, R-L, but I find the tracking really prefers people and cars across the street (so smaller objects?). This issue got worse when I moved picam720 60fps. Despite the wiki and very detailed information I'm struggling with what to adjust to make it want the larger closer cars moving vs. everything else. I calibrated for vehicle in mm and pixel and set a speed rule 19mph, but my signal not noise ratio is still way to high http://speed-car:8080/images/ car across http://speed-car:8080/images/ person L-R http://speed-car:8080/images/ leafs (ARGH)

I'm 100% sure this is just me not grasping what effects what in the functions.

[pageauc]

Upgrade speed camera to the latest version using menubox.sh UPGRADE menu pick

You can use the config.py setting to set one or both of the variables below. These can be changed to tune the system.

MIN_AREA = 200

and or

max_speed_over = 10 Monitor log activity to see what pedestrians average square pixels value and speed are. Set MIN_SREA = value and or max_speed_over variables close to log values.

You can also view historical data by enabling csv logging.

in config.py set variable below

log_data_to_CSV = True

Runs speed camera for a while to collect data and update the speed-cam.csv

Then run the ./makehtml.py script. This can also be run from menuubox.sh by running the HTML menu pick and viewing results in the web browser interface under html directory Note this report defaults to last 100 records but can be increased by editing the variable HTML_MAX_FILES = 100 in the makehtml.py script Note I am looking at putting makehtml.py settings in config.py

IMPORTANT: Try not to make the MIN_AREA variable too large since you may eliminate some vehicles that are tracked with smaller contours due to lighting or style. Eg sometimes a wheel, trunk or other smaller areas can be tracked.

Give that a try

Also I have found I get best results by using a smaller stream image size that reduces the amount of opencv processing. Default is 320x240 for webcam and pi-camera

Claude ...

On Thu, Jul 1, 2021 at 5:35 PM bushibot @.***> wrote:

I wonder if there might be a better place to ask? but I'm having some tuning configuration challenges. I need to catch vehicles moving L-R, R-L, but I find the tracking really prefers people and cars across the street (so smaller objects?). This issue got worse when I moved picam720 60fps. Despite the wiki and very detailed information I'm struggling with what to adjust to make it want the larger closer cars moving vs. everything else. I calibrated for vehicle in mm and pixel and set a speed rule 19mph, but my signal not noise ratio is still way to high http://speed-car:8080/images/ car across http://speed-car:8080/images/ person L-R http://speed-car:8080/images/ leafs (ARGH)

I'm 100% sure this is just me not grasping what effects what in the functions.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/pageauc/speed-camera/issues/94, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABNPKZHFDEAPVUQ4FCX7OCLTVTNRVANCNFSM47VQG4BA .

-- YouTube Channel at https://www.youtube.com/user/pageaucp http://www.youtube.com/user/pageaucp GitHub Repository at https://github.com/pageauc

[pageauc]

Added information to wiki https://github.com/pageauc/speed-camera/wiki/How-to-Tune-Motion-Tracking-Settings#tune-out-smaller-moving-objects

[bushibot]

Upgrade speed camera to the latest version using menubox.sh UPGRADE menu pick You can use the config.py setting to set one or both of the variables below. These can be changed to tune the system. MIN_AREA = 200 and or max_speed_over = 10 Monitor log activity to see what pedestrians average square pixels value and speed are. Set MIN_SREA = value and or max_speed_over variables close to log values. You can also view historical data by enabling csv logging. in config.py set variable below log_data_to_CSV = True Runs speed camera for a while to collect data and update the speed-cam.csv Then run the ./makehtml.py script. This can also be run from menuubox.sh by running the HTML menu pick and viewing results in the web browser interface under html directory Note this report defaults to last 100 records but can be increased by editing the variable HTML_MAX_FILES = 100 in the makehtml.py script Note I am looking at putting makehtml.py settings in config.py IMPORTANT: Try not to make the MIN_AREA variable too large since you may eliminate some vehicles that are tracked with smaller contours due to lighting or style. Eg sometimes a wheel, trunk or other smaller areas can be tracked. Give that a try Also I have found I get best results by using a smaller stream image size that reduces the amount of opencv processing. Default is 320x240 for webcam and pi-camera Claude ... … Thank you. I messed my setting up enough that I stated over. I could watch a car go through the tracking area and nothing would get locked 😤😂. Yes, oddly lower res seemed to work better for me too. Not as good for identifying the car though. Upgraded. Could I trouble you expand slightly on how the functions behave by example. If I shrink or grow the MIN_AREA, how dose that effect the tracking? Helping or hurting the contour it is looking for? How is min area effected by resolution, is there a scale consideration? I tried making a much larger car sized block but that failed to get cars but would still get bikes across the street. Speed is pretty clear, but a lot of the other variables are not, at least to me, as far as cause and effect.

Amazing when it it’s working.

[pageauc]

At approx line 1180 in speed-cam.py per code below is the only real programming spot (others are just for info) that MIN_AREA is used.. Moving objects can] be made of more than one contour due lighting, reflection, shape, Etc You can see the number of contours in the tracking logs when running speed-cam.py from a console with verbose turned on. In the log you will see C= that shows how many contours were found. Ideally this should be one. If the moving object is too close it can generate multiple contours. At any rate the program cycles through the contours and checks initially if it is larger than the MIN_AREA otherwise the contour is ignored. It then looks for the largest moving contour and that is the one it tracks. Determines the center x,y position of the contour center. That determines a single point..and a track is made up of a series of points that get filtered to make sure they are reasonably close to the previous point otherwise it is considered OUT of range and rejected. This can happen for lots of reasons like more than one object moving or the location of the largest contour shifting position on the moving object. This is what can cause erratic speed shifts especially if the camera is too close so that the moving object takes up a large part of the camera view. You will get this with a camera too close or things like trucks or like myself cutting the grass or birds flying close to camera and inside the tracking window. At any rate motion tracking follows the moving object where each point has to be within a certain number of pixels from the previous position. Contour movements on the object can be smoothed out over a longer track.

So MIN_AREA rejects all contours less than the setting. Most of the problems users have is the camera is less than approx 15m or 50f from the calibration zone or they try to set the camera stream resolution too high. Hope this helps.

    if contours
        total_contours = len(contours)
        motion_found = False
        biggest_area = MIN_AREA
        for c in contours:
            # get area of contour
            found_area = cv2.contourArea(c)
            if found_area > biggest_area:
                (x, y, w, h) = cv2.boundingRect(c)
                # check if object contour is completely within crop

area if (x > x_buf and x + w < x_right - x_left - x_buf):

                    track_x = x
                    track_y = y
                    track_w = w  # movement width of object contour
                    track_h = h  # movement height of object contour
                    motion_found = True
                    biggest_area = found_area

                 cur_track_time = time.time() # record cur track time

On Fri, Jul 2, 2021 at 3:19 AM bushibot @.***> wrote:

Upgrade speed camera to the latest version using menubox.sh UPGRADE menu pick You can use the config.py setting to set one or both of the variables below. These can be changed to tune the system. MIN_AREA = 200 and or max_speed_over = 10 Monitor log activity to see what pedestrians average square pixels value and speed are. Set MIN_SREA = value and or max_speed_over variables close to log values. You can also view historical data by enabling csv logging. in config.py set variable below log_data_to_CSV = True Runs speed camera for a while to collect data and update the speed-cam.csv Then run the ./makehtml.py script. This can also be run from menuubox.sh by running the HTML menu pick and viewing results in the web browser interface under html directory Note this report defaults to last 100 records but can be increased by editing the variable HTML_MAX_FILES = 100 in the makehtml.py script Note I am looking at putting makehtml.py settings in config.py IMPORTANT: Try not to make the MIN_AREA variable too large since you may eliminate some vehicles that are tracked with smaller contours due to lighting or style. Eg sometimes a wheel, trunk or other smaller areas can be tracked. Give that a try Also I have found I get best results by using a smaller stream image size that reduces the amount of opencv processing. Default is 320x240 for webcam and pi-camera Claude ... … <#m5174663356708830209> Thank you. I messed my setting up enough that I stated over. I could watch a car go through the tracking area and nothing would get locked 😤😂. Yes, oddly lower res seemed to work better for me too. Not as good for identifying the car though. Upgraded. Could I trouble you expand slightly on how the functions behave by example. If I shrink or grow the MIN_AREA, how dose that effect the tracking? Helping or hurting the contour it is looking for? I tried making a much larger car sized block but that failed to get cars but would still get bikes across the street. Speed is pretty clear, but a lot of the other variables are not, at least to me, as far as cause and effect.

Amazing when it it’s working.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/pageauc/speed-camera/issues/94#issuecomment-872778705, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABNPKZFT74C7L6R3XEWQIQTTVVR7VANCNFSM47VQG4BA .

-- YouTube Channel at https://www.youtube.com/user/pageaucp http://www.youtube.com/user/pageaucp GitHub Repository at https://github.com/pageauc