Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/billy-enrizky/video-exp-metrics

Extract  hand movements, timing, pipette angles, from a video
https://github.com/billy-enrizky/video-exp-metrics

Last synced: 3 months ago
JSON representation

Extract  hand movements, timing, pipette angles, from a video

Awesome Lists containing this project

README 

          
# Video-Based Pipetting Technique Analysis Metrics



This repository contains a comprehensive video analysis system for evaluating pipetting techniques in liquid handling experiments. The system analyzes operator performance through computer vision and generates quantitative metrics for reproducibility assessment.



See: `reproducibility_results_long/experienced_lab_worker_C0045/protocol_plots`



## Protocol Overview



**Objective**: To evaluate and compare the accuracy, reproducibility, and time efficiency of liquid dispensing between automated liquid handlers and manual pipetting across different operator skill levels.



**Complete Experimental Context**: This analysis system was developed for a comprehensive liquid handling study comparing manual pipetting across various people.



### Detailed Protocol Description



**Experiment 1: Basic Liquid Dispensing Protocol**

- **Protocol**: 4 volumes (50, 100, 150, 200 μL) × 4 liquids = 16 total dispensing events

- **Equipment**: Single-channel P200 pipette, P200 tips, 96-well plates, precision timer

- **Execution**: Start timer → complete all dispensing → stop timer (typically 15-25 minutes)



**The Four Liquid Types and Their Challenges**:



1. **DYE ddH₂O (Standard water with dye)**:

   - **Purpose**: Baseline measurement with visible liquid for accuracy verification

   - **Technique**: Standard pipetting (aspirate to first stop, dispense to first stop)

   - **Volumes**: 50, 100, 150, 200 μL dispensed 3 times each = 12 dispensing events

   - **Challenge Level**: Easy - standard aqueous solution



2. **DYE-FREE ddH₂O (Pure water)**:

   - **Purpose**: Control for dye interference, tests technique without visual aid

   - **Technique**: Standard pipetting

   - **Volumes**: 50, 100, 150, 200 μL dispensed once each = 4 dispensing events

   - **Challenge Level**: Moderate - harder to see dispensed volume



3. **GLYCEROL (Viscous liquid)**:

   - **Purpose**: Tests handling of high-viscosity liquids (glycerol ~1000× more viscous than water)

   - **Technique**: **Reverse pipetting required**

   - **Volumes**: 50, 100, 150, 200 μL dispensed once each = 4 dispensing events

   - **Challenge Level**: High - requires specialized technique



4. **ETHANOL (Volatile liquid)**:

   - **Purpose**: Tests handling of volatile, low-surface-tension liquids

   - **Technique**: **Pre-wetting required**

   - **Volumes**: 50, 100, 150, 200 μL dispensed once each = 4 dispensing events

   - **Challenge Level**: High - prone to dripping and evaporation



### Specialized Pipetting Techniques Explained



#### Reverse Pipetting (Required for Glycerol)

**Why Needed**: Glycerol's high viscosity (1000× water) causes liquid to stick to tip walls, leading to under-dispensing with standard technique.



**Technique Steps**:

1. Press pipette button to **second stop** (blow-out position)

2. Insert tip into liquid and **aspirate** (liquid fills tip plus extra air space)

3. Move to target location

4. Press button only to **first stop** to dispense (leaves residual liquid in tip)

5. **Extra movements**: Additional aspiration and partial dispensing create complex movement patterns



**Movement Pattern Impact**: 

- **Standard technique**: ~8 direction changes (down-aspirate-up-down-dispense-up)

- **Reverse technique**: ~12 direction changes (extra aspiration steps add 4 more movements)



#### Pre-wetting (Required for Ethanol)

**Why Needed**: Ethanol's high vapor pressure causes air expansion in pipette, leading to tip dripping and volume inaccuracy.



**Technique Steps**:

1. **Initial cycles**: Aspirate and dispense ethanol 2-3 times to saturate tip interior

2. **Purpose**: Equilibrate air space with ethanol vapor to prevent expansion

3. **Then proceed**: With normal aspiration and dispensing

4. **Extra movements**: Pre-wetting cycles double the movement patterns



**Movement Pattern Impact**:

- **Standard technique**: ~8 direction changes (single aspiration-dispensing cycle)

- **Pre-wetting technique**: ~12 direction changes (extra pre-wetting cycles add 4 more movements)



### Protocol Execution Flow and Timing



**Cycle Structure**: Each "cycle" represents handling one complete liquid type through all four volumes:

- **Cycle 1**: DYE ddH₂O (50→100→150→200 μL, 3× each = 12 events)

- **Cycle 2**: DYE-FREE ddH₂O (50→100→150→200 μL, 1× each = 4 events)  

- **Cycle 3**: GLYCEROL (50→100→150→200 μL, 1× each = 4 events)

- **Cycle 4**: ETHANOL (50→100→150→200 μL, 1× each = 4 events)



**Contamination Prevention**: Fresh tip required between each liquid type (4 tip changes total)



**Expected Timing**:

- **Expert operators**: 15-18 minutes (consistent pacing, smooth technique execution)

- **Trained students**: 18-22 minutes (following protocol carefully, some hesitation)

- **Novice operators**: 22-30 minutes (frequent pauses, technique uncertainties)



## Comprehensive Metrics Documentation



### Video Analysis Coordinate System

**Image Coordinate Convention**: 

- Origin (0,0) is at top-left corner of video frame

- X-axis increases from left to right

- Y-axis increases from top to bottom (downward positive)

- Hand positions tracked as (x,y) pixel coordinates of hand center

- Movement calculations use consecutive frame differences



**Movement Magnitude Context**:

- **Typical video resolution**: 1920×1080 pixels (Full HD)

- **Hand detection area**: Usually 200-800 pixels wide depending on camera angle

- **Actual movement scale**: 15 pixels ≈ 2-5mm real-world movement (varies with camera distance)

- **Frame rate**: Typically 30 FPS, so measurements are per 1/30th second intervals



### 1. Protocol Event Detection Metrics



#### 1.1 Aspiration Event Detection

**Formula**: `total_aspiration_events = count(upward_movement_magnitude > 15 AND confidence > 0.8)`

- Where `upward_movement_magnitude = sqrt(dx² + dy²)` when `dy < 0` (upward in image coordinates)

- Units: **pixels per frame**



**English Explanation**: Counts upward hand movements above threshold that indicate liquid aspiration into the pipette tip.



**Movement Magnitude Calculation**:

- `dx = current_hand_x - previous_hand_x` (horizontal displacement)

- `dy = current_hand_y - previous_hand_y` (vertical displacement, negative = upward)

- `upward_movement_magnitude = sqrt(dx² + dy²)` only when `dy < 0`

- **Units**: Measured in pixels per frame, representing the Euclidean distance the hand center moved between consecutive video frames



**Reasoning**: 

- **Threshold 15 pixels/frame**: Based on empirical analysis of pipetting videos where typical aspiration movements show 10-30 pixel displacements in standard video resolutions (1920×1080). A 15-pixel threshold captures deliberate pipetting motions while filtering out hand tremor (typically <5 pixels) and camera shake. This corresponds to approximately 2-5mm of actual hand movement depending on camera distance and zoom level.

- **Confidence 0.8**: High confidence threshold ensures we only count genuine hand movements, reducing false positives from detection noise

- **Directional filtering**: Only upward movements (dy < 0) are counted since aspiration requires lifting the pipette from liquid surface



#### 1.2 Dispensing Event Detection

**Formula**: `total_dispensing_events = count(downward_movement_magnitude > 15 AND confidence > 0.8)`

- Where `downward_movement_magnitude = sqrt(dx² + dy²)` when `dy > 0` (downward in image coordinates)

- Units: **pixels per frame**



**English Explanation**: Counts downward hand movements that indicate liquid dispensing from the pipette.



**Movement Magnitude Calculation**:

- `dx = current_hand_x - previous_hand_x` (horizontal displacement)

- `dy = current_hand_y - previous_hand_y` (vertical displacement, positive = downward)

- `downward_movement_magnitude = sqrt(dx² + dy²)` only when `dy > 0`

- **Units**: Measured in pixels per frame, representing the Euclidean distance the hand center moved between consecutive video frames



**Reasoning**: 

- **Threshold 15 pixels/frame**: Same empirical basis as aspiration detection. Dispensing movements typically show similar magnitude displacements as the operator moves the pipette tip to the target well and depresses the plunger. The 15-pixel threshold effectively distinguishes between purposeful dispensing motions and minor positioning adjustments.

- **Confidence 0.8**: High confidence threshold ensures we only count genuine hand movements, reducing false positives from detection noise

- **Directional filtering**: Only downward movements (dy > 0) are counted since dispensing requires lowering the pipette toward the target container



#### 1.3 Tip Change Event Detection

**Formula**: `tip_change_events = count(velocity < mean_velocity * 0.05 for extended_period)`

- Where `velocity = sqrt(dx² + dy²)` (movement magnitude regardless of direction)

- Units: **pixels per frame**



**English Explanation**: Detects periods of minimal hand movement indicating tip disposal and replacement.



**Velocity Calculation**:

- `velocity = sqrt((current_x - previous_x)² + (current_y - previous_y)²)`

- **Units**: Pixels per frame, representing total hand displacement between consecutive frames

- **Extended period**: Continuous low-velocity frames lasting >2 seconds (>60 frames at 30 FPS)



**Reasoning**: 

- **5% of mean velocity**: Very low activity threshold indicates hands are stationary during tip changes. If mean velocity is 20 pixels/frame, then 5% = 1 pixel/frame represents near-stillness

- **Extended period requirement**: Prevents counting brief pauses as tip changes while capturing actual equipment manipulation phases

- **Biological basis**: Tip changes require 3-5 seconds of careful manipulation with minimal hand movement



### 2. Timing Consistency Metrics



#### 2.1 Cycle Consistency Score

**Formula**: `cycle_consistency_score = 1 - (std(cycle_times) / mean(cycle_times))`



**English Explanation**: Measures how consistent the operator is in timing between repeated pipetting cycles. Higher scores indicate better reproducibility.



**Cycle Duration Definition**:

- **What is a "cycle"**: Complete handling of one liquid type through all four volumes (50, 100, 150, 200 μL)

- **Cycle boundaries**: From first interaction with a liquid type to tip disposal before next liquid

- **Timing calculation**: `cycle_duration = time_at_tip_disposal - time_at_first_aspiration`



**Example Cycle Breakdown**:

- **DYE ddH₂O cycle**: ~8-12 minutes (12 dispensing events: 3× each volume)

- **DYE-FREE ddH₂O cycle**: ~3-5 minutes (4 dispensing events: 1× each volume)

- **GLYCEROL cycle**: ~4-7 minutes (4 dispensing events + reverse pipetting complexity)

- **ETHANOL cycle**: ~4-7 minutes (4 dispensing events + pre-wetting steps)



**Cycle Time Variability Examples**:

- **Expert**: [480s, 240s, 300s, 280s] → std = 105s → consistency = 0.69 (good)

- **Student**: [600s, 300s, 420s, 380s] → std = 127s → consistency = 0.71 (good)

- **Novice**: [720s, 480s, 600s, 540s] → std = 102s → consistency = 0.72 (surprisingly good timing despite slower pace)



**Reasoning**: 

- **Coefficient of Variation (CV) inversion**: CV measures relative variability; subtracting from 1 makes higher values better

- **Range 0-1**: Provides intuitive scoring where 1.0 = perfect consistency, 0.0 = highly variable

- **Skill assessment**: Experienced operators show more consistent cycle-to-cycle timing regardless of liquid complexity



#### 2.2 Time Efficiency Score

**Formula**: `time_efficiency_score = expected_completion_time / actual_total_time`



**English Explanation**: Compares actual protocol completion time against expected duration (20 minutes = 1200 seconds).



**Reasoning**: 

- **Expected time 1200s (20 min)**: Based on protocol design and expert performance benchmarks

- **Ratio format**: Values >1.0 indicate faster than expected, <1.0 indicate slower than expected



### 3. Technique-Specific Adherence Metrics



#### 3.1 Reverse Pipetting Adherence (Glycerol)

**Formula**: `reverse_pipetting_score = min(1.0, detected_extra_steps / expected_extra_steps)`

- Where `expected_extra_steps = 2` and `expected_direction_changes = 12`



**English Explanation**: Measures adherence to reverse pipetting technique required for viscous liquids like glycerol.



**Direction Changes Detailed Breakdown**:



**Standard Pipetting Technique (8 direction changes)**:

1. **Down**: Lower pipette to liquid surface

2. **Aspirate**: Press button to first stop, release to aspirate

3. **Up**: Lift pipette from liquid  

4. **Move**: Horizontal movement to target well

5. **Down**: Lower pipette to target well

6. **Dispense**: Press button to first stop to dispense

7. **Up**: Lift pipette from well

8. **Move**: Return to rest position



**Reverse Pipetting Technique (12 direction changes)**:

1. **Down**: Lower pipette to liquid surface

2. **Pre-aspirate**: Press button to **second stop** (blow-out position)

3. **Aspirate**: Release button to aspirate extra volume

4. **Up**: Lift pipette from liquid

5. **Move**: Horizontal movement to target well  

6. **Down**: Lower pipette to target well

7. **Partial dispense**: Press button only to **first stop** (retains residual)

8. **Up**: Lift pipette from well

9. **Evaluate**: Visual check of dispensed volume

10. **Adjust**: Possible additional small dispense if needed

11. **Final up**: Complete withdrawal

12. **Move**: Return to rest position



**Reasoning**: 

- **Extra steps = 2**: Reverse pipetting requires additional aspiration beyond first stop, creating 2 extra movement phases (pre-aspirate + residual management)

- **Direction changes = 12**: More movement reversals due to the complex aspiration pattern vs. 8 for standard technique

- **Technique necessity**: Glycerol's high viscosity (η = 1.41 Pa·s vs water's 0.001 Pa·s) requires this specialized technique for accuracy

- **Detection method**: Algorithm counts velocity direction reversals to identify technique complexity



#### 3.2 Pre-wetting Adherence (Ethanol)

**Formula**: `pre_wetting_score = 1.0 if direction_changes >= 12 else 0.7 if >= 8 else 0.3`



**English Explanation**: Assesses whether the operator performed pre-wetting cycles before ethanol aspiration to prevent tip dripping.



**Direction Changes Detailed Breakdown**:



**Standard Pipetting Technique (8 direction changes)**:

1. **Down**: Lower pipette to liquid surface

2. **Aspirate**: Press button, release to aspirate

3. **Up**: Lift pipette from liquid

4. **Move**: Horizontal movement to target

5. **Down**: Lower pipette to target well

6. **Dispense**: Press button to dispense

7. **Up**: Lift pipette from well

8. **Move**: Return to rest position



**Pre-wetting Technique (12+ direction changes)**:

1. **Down**: Lower pipette to ethanol surface

2. **Pre-wet 1**: Aspirate small volume

3. **Pre-dispense 1**: Dispense back into source (equilibrates tip)

4. **Pre-wet 2**: Second aspiration cycle 

5. **Pre-dispense 2**: Second dispense back into source

6. **Final aspirate**: Aspirate target volume

7. **Up**: Lift pipette from liquid

8. **Move**: Horizontal movement to target

9. **Down**: Lower pipette to target well

10. **Dispense**: Press button to dispense

11. **Up**: Lift pipette from well

12. **Move**: Return to rest position



**Reasoning**: 

- **12 direction changes**: Pre-wetting requires initial aspiration-dispense cycles, doubling normal movement patterns

- **Graduated scoring**: Accounts for partial adherence (some pre-wetting) vs. complete adherence

- **Ethanol volatility**: High vapor pressure (5.95 kPa at 20°C vs water's 2.34 kPa) causes tip dripping without pre-wetting, affecting accuracy

- **Physics basis**: Ethanol vapor expands air cushion in pipette, causing uncontrolled liquid expulsion



### 4. Spatial Movement Metrics



#### 4.1 Spatial Variability

**Formula**: `spatial_variability = std(distances_from_center)`

- Where `distances_from_center = sqrt((x - center_x)² + (y - center_y)²)`

- Units: **pixels** (standard deviation of distances)



**English Explanation**: Measures how much hand position varies from the average working position during protocol execution.



**Distance Calculation**:

- `center_x = mean(all_hand_x_positions)` across entire protocol

- `center_y = mean(all_hand_y_positions)` across entire protocol  

- `distance_i = sqrt((x_i - center_x)² + (y_i - center_y)²)` for each frame i

- `spatial_variability = standard_deviation(all_distances)`



**Reasoning**: 

- **Euclidean distance**: Standard geometric measure of position deviation

- **Center-based**: Uses mean position as reference to account for individual working preferences

- **Lower values**: Indicate more consistent, controlled movements (typical expert: <30 pixels, novice: >80 pixels)

- **Units context**: 30 pixels ≈ 3-8mm real-world spatial consistency depending on camera setup



#### 4.2 Hand Steadiness Score

**Formula**: `steadiness_score = 1.0 - (velocity_std / (velocity_mean + 0.001))`

- Where `velocity = sqrt(dx² + dy²)` for each frame transition

- Units: **dimensionless score** (0-1 scale)



**English Explanation**: Quantifies hand stability during pipetting operations, with higher scores indicating steadier hands.



**Velocity Statistics**:

- `velocity_mean = average(all_frame_velocities)` in pixels/frame

- `velocity_std = standard_deviation(all_frame_velocities)` in pixels/frame

- **Coefficient of Variation**: `CV = velocity_std / velocity_mean` (dimensionless)

- **Steadiness Score**: `1.0 - CV` (inverted so higher = steadier)



**Reasoning**: 

- **CV-based measure**: Relative variability accounts for different movement speeds across operators

- **Epsilon term (+0.001)**: Prevents division by zero in perfectly still moments (rare but possible)

- **Inversion (1.0 -)**: Makes higher values represent better steadiness (0.9+ = very steady, <0.5 = shaky)

- **Biological interpretation**: Reflects natural hand tremor, fatigue, and motor control precision



### 5. Volume-Specific Consistency Metrics



#### 5.1 Duration Consistency CV

**Formula**: `duration_cv = (std(cycle_durations) / mean(cycle_durations)) * 100`



**English Explanation**: Coefficient of variation for time taken to complete each volume dispensing cycle.



**Reasoning**: 

- **Percentage format**: Standard CV presentation for easy interpretation

- **Per-volume analysis**: Different volumes require different timing, so consistency is measured within each volume group

- **Reproducibility indicator**: Lower CV indicates more reproducible technique



#### 5.2 Velocity Consistency CV

**Formula**: `velocity_cv = (std(avg_velocities) / mean(avg_velocities)) * 100`



**English Explanation**: Measures consistency in movement speed across cycles for each volume.



**Reasoning**: 

- **Movement speed**: Different volumes may require different speeds, but consistency within volume is crucial

- **Technical skill indicator**: Experienced operators maintain consistent velocities for each volume



### 6. Accuracy Prediction Metrics



#### 6.1 Predicted CV Percentage

**Formula**: `predicted_cv = (velocity_cv * 0.6 + duration_cv * 0.4) / 5.0`



**English Explanation**: Predicts the coefficient of variation in actual dispensed volumes based on movement consistency.



**Reasoning**: 

- **Velocity weight 0.6**: Movement consistency is primary predictor of pipetting accuracy

- **Duration weight 0.4**: Timing consistency contributes but is secondary to movement control

- **Scale factor 5.0**: Empirical scaling to match typical pipetting CV ranges (1-10%)



#### 6.2 Accuracy Target Assessment

**Formula**: `meets_target = predicted_cv <= target_cv`

- Target CVs: 50μL < 5%, 100μL < 3%, 150μL < 2.5%, 200μL < 2%



**English Explanation**: Determines if predicted performance meets volume-specific accuracy standards.



**Reasoning**: 

- **Volume-dependent targets**: Larger volumes typically achieve better relative precision

- **Industry standards**: Based on published pipetting accuracy requirements for laboratory work

- **50μL target 5%**: Smallest volume has highest relative error tolerance

- **200μL target 2%**: Largest volume should achieve highest relative precision



### 7. Operator Classification Metrics



#### 7.1 Overall Reproducibility Score

**Formula**: `overall_score = mean([timing_score, spatial_score, velocity_score])`

Where:

- `timing_score = 1.0 - min(timing_variability / 10.0, 1.0)`

- `spatial_score = 1.0 - min(spatial_variability / 100.0, 1.0)`

- `velocity_score = 1.0 - min(velocity_variability / 50.0, 1.0)`



**English Explanation**: Composite score combining timing, spatial, and velocity consistency into overall reproducibility assessment.



**Component Variability Definitions**:



**Timing Variability**:

- **Formula**: `timing_variability = std(cycle_durations)` 

- **Units**: **seconds** (standard deviation of cycle completion times)

- **Calculation**: Protocol divided into 4 cycles (one per liquid), duration measured for each cycle

- **Example**: If cycle times are [45s, 50s, 48s, 52s], then timing_variability = std([45,50,48,52]) = 2.9 seconds

- **Interpretation**: Lower values indicate more consistent pacing across protocol cycles



**Cycle Duration Context by Liquid Type**:

- **DYE ddH₂O cycle**: Longest duration (8-12 min) due to 12 dispensing events (3× each volume)

- **Other cycles**: Shorter duration (3-7 min) due to 4 dispensing events each

- **Complexity impact**: GLYCEROL and ETHANOL cycles often take longer due to specialized techniques

- **Skill differentiation**: Experts maintain consistent timing regardless of liquid complexity



**Real-world Examples**:

- **Expert operator**: [480s, 240s, 300s, 280s] → timing_variability = 105s (excellent consistency)

- **Trained student**: [600s, 300s, 420s, 380s] → timing_variability = 127s (good consistency)  

- **Novice operator**: [900s, 600s, 750s, 680s] → timing_variability = 123s (surprising consistency despite slower pace)



**Velocity Variability**: 

- **Formula**: `velocity_variability = coefficient_of_variation(cycle_velocity_patterns)`

- **Units**: **dimensionless** (relative variability in movement patterns)

- **Calculation**: 

  - Divide protocol into 4 cycles

  - Calculate mean and std velocity for each cycle

  - Measure variability in these statistics across cycles

  - `velocity_variability = (std_cv + mean_cv) / 2` where cv = coefficient of variation

- **Example**: If cycles show velocity CVs of [0.2, 0.25, 0.22, 0.28], velocity_variability reflects consistency of movement patterns

- **Interpretation**: Lower values indicate more repeatable movement dynamics across cycles



**Reasoning**: 

- **Normalization factors**: 

  - **10.0s timing**: Maximum acceptable cycle-to-cycle timing variation (expert: <2s, novice: >5s)

  - **100px spatial**: Maximum acceptable spatial inconsistency (expert: <30px, novice: >80px) 

  - **50px/frame velocity**: Maximum acceptable velocity pattern variation

- **Equal weighting**: All three components equally important for overall technique assessment

- **0-1 scale**: Provides intuitive scoring system where 1.0 = perfect reproducibility, 0.0 = highly variable



#### 7.2 Operator Type Classification

**Formula**: 

```

if reproducibility > 0.9 AND timing_consistency > 0.9:

    return "automated_liquid_handler"

elif technique_adherence > 0.8 AND timing_consistency > 0.7:

    return "experienced_lab_worker"  

elif technique_adherence > 0.6 AND reproducibility > 0.5:

    return "freshly_trained_student"

else:

    return "novice_learning"

```



**English Explanation**: Classifies operator skill level based on consistency and technique adherence thresholds.



**Reasoning**: 

- **Automated handler thresholds (0.9, 0.9)**: Machines should show near-perfect consistency

- **Experienced worker thresholds (0.8, 0.7)**: High technique knowledge with good consistency

- **Trained student thresholds (0.6, 0.5)**: Follows protocols but with moderate consistency

- **Progressive thresholds**: Reflect realistic performance expectations for each skill level



### 8. Contamination Prevention Metrics



#### 8.1 Cross-Contamination Prevention Score

**Formula**: `contamination_score = min(1.0, low_activity_periods / (total_periods * 0.1))`



**English Explanation**: Measures frequency of low-activity periods indicating proper contamination prevention procedures.



**Reasoning**: 

- **10% target**: Expect ~10% of time spent on contamination prevention activities

- **Low activity detection**: Periods of minimal movement indicate tip changes, cleaning, or waiting

- **Protocol requirement**: Essential for multi-liquid protocols to prevent cross-contamination



#### 8.2 Tip Change Technique Score

**Formula**: `tip_change_score = min(1.0, detected_tip_changes / expected_tip_changes)`

- Where `expected_tip_changes = 4` (one per liquid type)



**English Explanation**: Assesses whether operator changed tips appropriately between different liquids.



**Reasoning**: 

- **4 expected changes**: Protocol requires fresh tip for each liquid (DYE ddH₂O, DYE-FREE ddH₂O, GLYCEROL, ETHANOL)

- **Detection method**: Extended periods of minimal movement followed by resumed activity

- **Contamination prevention**: Critical for maintaining liquid purity and experimental validity



### 9. Hand Detection Technical Metrics



#### 9.1 Hand Detection Rate

**Formula**: `detection_rate = frames_with_hands / total_processed_frames`



**English Explanation**: Percentage of video frames where hands were successfully detected by computer vision system.



**Reasoning**: 

- **Quality indicator**: Higher detection rates indicate better video quality and algorithm performance

- **Confidence threshold**: Uses detection confidence > 0.2 to balance sensitivity vs. false positives

- **Glove compatibility**: Adjusted thresholds for gloved hand detection scenarios



#### 9.2 Detection Confidence Score

**Formula**: `avg_confidence = mean(detection_confidence_scores)`



**English Explanation**: Average confidence level of hand detection across all frames.



**Reasoning**: 

- **Algorithm reliability**: Higher confidence indicates more reliable tracking

- **Range 0-1**: MediaPipe confidence scores provide quality assessment

- **Threshold selection**: 0.2 minimum chosen to include gloved hands while excluding noise



## Analysis Pipeline Summary



The complete analysis pipeline processes video data through:



1. **Frame-by-frame hand detection** with glove-optimized parameters

2. **Movement pattern analysis** to identify protocol events

3. **Timing analysis** to measure cycle consistency

4. **Technique assessment** for specialized liquid handling

5. **Statistical analysis** to predict accuracy and classify operators

6. **Comprehensive reporting** with visual analytics



Each metric is designed to capture specific aspects of pipetting technique that correlate with actual laboratory performance, enabling objective assessment of operator skill and protocol adherence.