Color detection is the foundation of PhysisLab’s camera tracking system. This guide teaches you how to select, calibrate, and fine-tune HSV color ranges for robust object detection under varying lighting conditions.
Why HSV Color Space? HSV (Hue, Saturation, Value) separates color information from brightness, making it superior to RGB for color-based tracking:
Hue (0-179): The actual color (red, green, blue, etc.)Saturation (0-255): Color intensity (vivid vs. pale)Value (0-255): Brightness (light vs. dark)
OpenCV uses H: 0-179 (not 0-360) to fit in 8 bits. When converting from other tools, divide Hue by 2.
Basic Color Calibration Method 1: ROI-Based Auto-Calibration The simplest approach is to select a region of the object and calculate mean HSV values:
import cv2 import numpy as np # Capture reference frame ret, frame = cap.read() # Select ROI around the object roi = cv2.selectROI( "Selecciona region del objeto" , frame, False , False ) x, y, w, h = roi # Extract and convert to HSV selected_region = frame[y:y + h, x:x + w] hsv_region = cv2.cvtColor(selected_region, cv2. COLOR_BGR2HSV ) # Calculate mean HSV mean_hsv = np.mean(hsv_region.reshape( - 1 , 3 ), axis = 0 ).astype( int ) print ( f "HSV promedio: H= { mean_hsv[ 0 ] } , S= { mean_hsv[ 1 ] } , V= { mean_hsv[ 2 ] } " )
Setting Tolerance Values Fixed Tolerance
Adaptive Tolerance
# Simple approach with fixed tolerance tolerance = np.array([ 25 , 85 , 85 ]) # [H, S, V] lower_color = np.clip(mean_hsv - tolerance, 0 , 255 ) upper_color = np.clip(mean_hsv + tolerance, 0 , 255 ) # Note: H is 0-179, so upper bound should be 179 lower_color[ 0 ] = max ( 0 , mean_hsv[ 0 ] - tolerance[ 0 ]) upper_color[ 0 ] = min ( 179 , mean_hsv[ 0 ] + tolerance[ 0 ])
Advanced Calibration Techniques Handling Low Saturation Objects For pale or white objects (low saturation), widen the saturation range:
h_bob = np.mean(hsv_bob[:, :, 0 ]) s_bob = np.mean(hsv_bob[:, :, 1 ]) v_bob = np.mean(hsv_bob[:, :, 2 ]) margen_h = 15 margen_s = max ( 40 , s_bob * 0.4 ) # Wider range for low saturation margen_v = max ( 40 , v_bob * 0.4 ) hsv_bob_lower = np.array([ max ( 0 , h_bob - margen_h), max ( 0 , s_bob - margen_s), max ( 0 , v_bob - margen_v) ]) hsv_bob_upper = np.array([ min ( 179 , h_bob + margen_h), min ( 255 , s_bob + margen_s), min ( 255 , v_bob + margen_v) ])
Calibrating Multiple Objects For experiments with multiple tracked objects, calibrate each separately:
# Calibrate pendulum bob print ( "Selecciona ROI del CUERPO del péndulo (bob)" ) roi_bob = cv2.selectROI( "Seleccionar BOB" , frame_ref, False ) xb, yb, wb, hb = roi_bob objeto_bob = frame_ref[yb:yb + hb, xb:xb + wb] hsv_bob = cv2.cvtColor(objeto_bob, cv2. COLOR_BGR2HSV ) # ... calculate bob color range ... # Calibrate pivot/axis print ( "Selecciona ROI del EJE/PIVOTE del péndulo" ) roi_eje = cv2.selectROI( "Seleccionar EJE" , frame_ref, False ) xe, ye, we, he = roi_eje objeto_eje = frame_ref[ye:ye + he, xe:xe + we] hsv_eje = cv2.cvtColor(objeto_eje, cv2. COLOR_BGR2HSV ) # ... calculate pivot color range ...
Applying Color Masks Basic Masking Once you have color ranges, apply the mask to each frame:
while True : ret, frame = cap.read() if not ret: break # Convert to HSV hsv = cv2.cvtColor(frame, cv2. COLOR_BGR2HSV ) # Apply color threshold mask = cv2.inRange(hsv, lower_color, upper_color) # Show mask for debugging cv2.imshow( "Mask" , mask) cv2.imshow( "Frame" , frame) if cv2.waitKey( 1 ) & 0x FF == 27 : break
Morphological Operations
Opening: Remove Noise
Morphological opening removes small isolated pixels: kernel = np.ones(( 5 , 5 ), np.uint8) mask = cv2.morphologyEx(mask, cv2. MORPH_OPEN , kernel)
This is erosion followed by dilation - it removes small white noise.
Dilation: Fill Gaps
Dilation fills small holes inside the detected object: mask = cv2.morphologyEx(mask, cv2. MORPH_DILATE , kernel)
Closing: Connect Regions (Optional)
For fragmented detections, closing can help: mask = cv2.morphologyEx(mask, cv2. MORPH_CLOSE , kernel)
This is dilation followed by erosion . Visualizing HSV Values Interactive Color Picker Create a tool to click on the object and see HSV values:
import cv2 import numpy as np def click_color ( event , x , y , flags , param ): if event == cv2. EVENT_LBUTTONDOWN : frame, hsv = param h, s, v = hsv[y, x] bgr = frame[y, x] print ( f "Pixel ( { x } , { y } ): HSV=( { h } , { s } , { v } ) BGR=( { bgr[ 0 ] } , { bgr[ 1 ] } , { bgr[ 2 ] } )" ) cap = cv2.VideoCapture( 0 ) ret, frame = cap.read() hsv = cv2.cvtColor(frame, cv2. COLOR_BGR2HSV ) cv2.namedWindow( "Frame" ) cv2.setMouseCallback( "Frame" , click_color, (frame, hsv)) cv2.imshow( "Frame" , frame) cv2.waitKey( 0 ) cv2.destroyAllWindows()
Histogram Analysis Visualize the distribution of HSV values in your ROI:
import matplotlib.pyplot as plt # After selecting ROI hsv_region = cv2.cvtColor(selected_region, cv2. COLOR_BGR2HSV ) fig, axs = plt.subplots( 1 , 3 , figsize = ( 12 , 4 )) for i, (channel, name) in enumerate ([( 0 , 'Hue' ), ( 1 , 'Saturation' ), ( 2 , 'Value' )]): axs[i].hist(hsv_region[:,:,channel].ravel(), bins = 50 , color = [ 'r' , 'g' , 'b' ][i]) axs[i].set_title(name) axs[i].set_xlabel( 'Value' ) axs[i].set_ylabel( 'Frequency' ) plt.tight_layout() plt.show()
Color Selection Tips
Red Objects Hue wraps around at 180! Use two masks:
Lower red: H=0-10
Upper red: H=170-179
mask1 = cv2.inRange(hsv, ( 0 , 100 , 100 ), ( 10 , 255 , 255 )) mask2 = cv2.inRange(hsv, ( 170 , 100 , 100 ), ( 179 , 255 , 255 )) mask = cv2.bitwise_or(mask1, mask2)
Yellow/Orange H: 15-35 Good for high-visibility objects. Wide saturation range recommended.
Green H: 40-80 Avoid if background is green (grass, walls). Use blue or yellow instead.
Blue H: 90-130 Excellent for indoor experiments with neutral backgrounds.
Troubleshooting Color Detection Problem: Object Not Detected
Print actual HSV values and compare with your thresholds: print ( f "Lower: { lower_color } " ) print ( f "Upper: { upper_color } " ) print ( f "Mean: { mean_hsv } " )
Always display the binary mask to see what’s being detected:
If the mask is empty, widen your ranges: tolerance = np.array([ 30 , 100 , 100 ]) # Wider
Problem: Too Many False Detections
Narrow HSV Range
Reduce tolerance, especially for Hue: margen_h = 10 # Instead of 15
Increase Morphological Kernel
Larger opening removes bigger noise: kernel = np.ones(( 7 , 7 ), np.uint8) # Instead of 5x5
Filter by Area
Reject contours that are too small or too large: if cv2.contourArea(c) > 300 and cv2.contourArea(c) < 5000 : # Process contour
Problem: Detection Lost During Motion Shadows and reflections change the object’s apparent color. Solutions:
Use diffuse lighting (softbox or bounced light)
Increase Value tolerance to handle brightness changes
Coat object with matte paint to reduce reflections
Use multiple color channels and combine masks
Lighting Best Practices Recommended Setup ┌─────────────┐ │ Diffuser │ ← Softbox or white sheet │ Light │ └──────┬──────┘ ↓ [Object] ← Your tracked object ↓ Camera
Do
Use indirect/diffuse lighting
Keep lighting consistent during experiment
Use high-contrast backgrounds
Test calibration before recording
Don't
Use direct sunlight (changes intensity)
Mix different light sources (color temperature)
Rely on auto white balance
Ignore shadows on the object
Real-World Examples Free Fall Experiment # Bright yellow ball against dark background tolerance = np.array([ 25 , 85 , 85 ]) # Typical HSV: (25, 200, 220) - Yellow/Orange
Pendulum Tracking # Red bob + gray pivot - two separate calibrations margen_h = 15 margen_s = max ( 40 , s_bob * 0.4 ) # Adaptive for varying saturation margen_v = max ( 40 , v_bob * 0.4 )
Mass-Spring System analisis.py (masa-resorte)
# Three green markers for reference triangle kernel_sz = 7 # Larger kernel for far-away markers n_esperados = 3 # Detect exactly 3 markers
Next Steps
Camera Tracking Apply color detection in motion tracking pipelines
Data Analysis Process tracking data to extract physics measurements