Hidden Hounds: Data-Driven Detects Furry Life Patterns

Introduction to Canine Behavior Analysis

Understanding what makes a dog behave the way it does is essential for trainers, veterinarians, and owners alike. Traditional observation methods—watching a dog in a kennel or during a walk—provide valuable insights but are limited by human perception and time constraints. Modern technology now allows us to capture objective data on a dog's movements, heart rate, vocalizations, and environmental interactions using wearable sensors.

Why Sensor Data Matters

• Continuous Monitoring: Sensors record thousands of data points per minute, revealing patterns that brief human observations miss.
• Quantifiable Metrics: Variables such as acceleration, temperature, and GPS coordinates can be converted into measurable indices (e.g., “activity index,” “resting heart rate”).
• Objective Baselines: By establishing a dog's typical behavior profile, deviations become easier to detect—useful for early warning of stress or illness.

Typical Sensors Used in Canine Studies

From Raw Data to Insightful Trends

1. Data Collection: Attach a lightweight collar or vest that houses the sensor suite. Ensure the device is comfortable and does not impede natural movement.
2. Pre‑Processing: Filter out noise (e.g., wind affecting microphones), correct for missing timestamps, and segment data into behavioral bouts using thresholding on acceleration.
3. Feature Extraction: Compute metrics such as mean daily activity, peak heart rate during play, or average distance traveled per hour.
4. Statistical Analysis: Apply time‑series models (ARIMA) or machine learning classifiers (Random Forests) to detect abnormal patterns.
5. Visualization & Reporting: Create dashboards that plot activity vs. heart rate over weeks, highlighting periods of stress or improvement after training interventions.

Practical Advice for Beginners

• Select the Right Device: For hobbyists, a commercial dog GPS collar with built‑in accelerometer often suffices. Researchers may need custom-built units with higher sampling rates.
• Start Small: Begin by monitoring one dog for two weeks to establish baseline behavior before adding more subjects.
• Ensure Ethical Use: Verify that the sensor weight is less than 5% of the dog's body mass and that data privacy regulations (e.g., GDPR) are respected when sharing location information.
• Integrate with Existing Software: Many platforms (e.g., Petcube, Garmin for pets) provide APIs. Use Python libraries like pandas and matplotlib to process and visualize the data.
• Iterative Refinement: After initial analysis, refine sensor placement or sampling frequency based on observed artifacts.

Case Study: Reducing Separation Anxiety with Sensor Feedback

A 4‑year‑old Labrador named Max exhibited excessive barking and pacing when left alone. By equipping him with a collar that recorded heart rate and GPS, researchers found:

• Heart Rate Spike: Max’s HR increased from 70 bpm (rest) to 120 bpm within the first minute of separation.
• Pacing Path: GPS logs showed he walked a 20‑meter loop around the house before settling.

Intervention: A weighted blanket and a scheduled “play session” after 30 minutes. Follow‑up data revealed heart rates dropping to 80 bpm and reduced pacing loops by 75%. This concrete, sensor‑derived evidence guided both training protocols and owner reassurance.

Conclusion

By harnessing sensor data, canine behavior analysis moves from subjective anecdote to quantifiable science. Whether you’re a seasoned researcher or a devoted pet parent, integrating wearable technology into your observational toolkit can unlock deeper understanding, more effective interventions, and ultimately healthier, happier dogs.

Types of Sensors for Dog Monitoring

When it comes to monitoring dogs, the variety of sensors available today is staggering. Each type brings its own set of data points that can be leveraged not only for immediate health or safety alerts but also for long‑term behavioral research. Below we break down the most common sensor categories, illustrate how they function in real‑world scenarios, and provide actionable tips on integrating them into a cohesive monitoring system.

1. Accelerometers & Gyroscopes (Inertial Measurement Units)

• What They Do: Detect linear acceleration, orientation, and angular velocity.
• Typical Use Cases:
  • Activity level tracking (e.g., distinguishing between walking, running, or resting).
  • Detecting falls or sudden jerks that may indicate injury.
  • Estimating energy expenditure for fitness and weight management.
• Practical Tip: Pair an IMU with a simple machine‑learning model to classify behaviors (e.g., “playful,” “aggressive,” “sleeping”). This can be done locally on the device or via cloud inference.

2. GPS & GNSS Modules

• What They Do: Provide real‑time location, speed, and movement history.
• Typical Use Cases:
  • Tracking wanderers or preventing escape incidents.
  • Mapping preferred routes for outdoor activities.
  • Analyzing spatial patterns of activity (e.g., time spent in certain zones).
• Practical Tip: Combine GPS data with accelerometer logs to correlate speed changes with behavioral states, revealing insights such as “the dog stops abruptly near the neighbor’s fence.”

3. Heart Rate Monitors (PPG or ECG)

• What They Do: Measure pulse rate and heart‑rate variability.
• Typical Use Cases:
  • Monitoring stress levels during training sessions.
  • Detecting arrhythmias or other cardiac issues early.
  • Assessing recovery after exercise.
• Practical Tip: Use HRV thresholds to flag potential anxiety episodes—useful for dogs that exhibit “pacing” behavior during thunderstorms.

4. Temperature Sensors

• What They Do: Record ambient or skin temperature.
• Typical Use Cases:
  • Preventing overheating in hot climates.
  • Detecting fever indicative of infection.
  • Monitoring microclimate changes inside a kennel.
• Practical Tip: Combine temperature data with GPS to identify “hot spots” in the environment where dogs may be at risk of heatstroke.

5. Microphone Arrays & Acoustic Sensors

• What They Do: Capture vocalizations, barking patterns, and ambient sounds.
• Typical Use Cases:
  • Analyzing bark frequency to differentiate between “alert” vs. “playful” barks.
  • Detecting distress calls (e.g., during separation anxiety).
  • Monitoring environmental noise that may affect behavior.
• Practical Tip: Deploy a lightweight audio classifier to flag abnormal vocalizations, then cross‑reference with accelerometer data for context.

6. Pressure & Load Sensors (e.g., Smart Collars)

• What They Do: Measure tension on the neck or weight distribution.
• Typical Use Cases:
  • Ensuring leash safety during walks.
  • Detecting signs of strain or injury in joints.
  • Monitoring posture changes that may indicate discomfort.
• Practical Tip: Set alerts when tension exceeds a safe threshold, especially during high‑intensity training sessions.

7. Environmental Sensors (Humidity, Air Quality, Light)

• What They Do: Provide context about the surrounding environment.
• Typical Use Cases:
  • Identifying environmental triggers for allergy flare‑ups.
  • Adjusting indoor climate control to keep dogs comfortable.
  • Correlating light levels with sleep quality.
• Practical Tip: Use an IoT hub to feed environmental data into a behavioral model, revealing patterns such as “the dog sleeps poorly when indoor humidity exceeds 70%.”

Integrating Multiple Sensors: A Practical Workflow

1. Data Collection: Deploy wearable or stationary devices that gather synchronized streams of data.
2. Pre‑processing: Clean and align timestamps, apply filters (e.g., low‑pass for accelerometer noise).
3. Feature Extraction: Derive meaningful metrics such as step count, average heart rate, GPS speed, bark duration.
4. Modeling: Train supervised classifiers or unsupervised clustering algorithms to identify behavior patterns.
5. Visualization & Alerts: Use dashboards (Grafana, Power BI) and mobile push notifications for real‑time insights.

By thoughtfully combining these sensor modalities, researchers can uncover nuanced trends in canine behavior—such as seasonal activity peaks, stress triggers during training, or the impact of environmental changes on sleep patterns. For pet owners, this integrated approach translates into proactive care and a deeper understanding of their dog’s daily life.

Setting Up a Sensor Network in the Home

# Step 1: Install sensors
# Step 2: Configure MQTT topics (e.g., /home/livingroom/motion)
# Step 3: Script to log data every second
# Step 4: Run R script to compute activity heatmap per day
# Step 5: Export results to CSV for statistical analysis

Data Collection Protocols and Ethics

Preprocessing Raw Sensor Data

The raw data collected from wearable devices, GPS trackers, or home-installed sensors are often noisy, inconsistent, and unstructured. Before you can feed this information into any analytical model—whether a simple trend‑analysis script or a complex machine learning pipeline—you must clean, transform, and enrich the dataset. Below is a step‑by‑step guide to preprocessing canine sensor data, complete with practical examples and code snippets.

1. Data Ingestion & Format Standardization

• File types: Sensors may export CSV, JSON, or proprietary binary formats. Use a uniform schema (e.g., timestamp in ISO8601, latitude/longitude for GPS).
• Timestamp alignment: Convert all timestamps to UTC and resample to a common frequency (e.g., 1 Hz for accelerometers, 1 min for GPS).
• Example (Python):

import pandas as pd

# Load CSV
df = pd.read_csv('accelerometer_raw.csv', parse_dates=['timestamp'])
df['timestamp'] = df['timestamp'].dt.tz_localize('UTC')
df.set_index('timestamp', inplace=True)

# Resample to 1Hz
df_resampled = df.resample('1S').mean()

2. Noise Filtering

• Low‑pass filters: Remove high‑frequency noise from accelerometer data using a Butterworth filter.
• Smoothing GPS trajectories: Apply a Kalman filter or moving‑average window to mitigate jitter.
• Example (SciPy):

from scipy.signal import butter, filtfilt

def lowpass_filter(data, cutoff=5, fs=50, order=4):
    nyq = 0.5 * fs
    normal_cutoff = cutoff / nyq
    b, a = butter(order, normal_cutoff, btype='low', analog=False)
    return filtfilt(b, a, data)

df_resampled['ax_clean'] = lowpass_filter(df_resampled['ax'])

3. Missing Data Handling

• Gap detection: Identify time windows with no sensor readings.
• Imputation strategies:
  • Forward/backward fill for short gaps (< 5 min).
  • Linear interpolation or spline for medium gaps (5–30 min).
  • Model‑based imputation (e.g., KNN, MICE) for larger gaps.
• Example:

df_resampled['ax_clean'].interpolate(method='time', limit=30, limit_direction='both')

4. Feature Extraction & Enrichment

• Activity classification: Use thresholds on acceleration magnitude to label periods as resting, walking, running.
• GPS speed & heading: Compute instantaneous speed and direction from latitude/longitude pairs.
• Environmental context: Append weather data (temperature, humidity) or indoor/outdoor flags via API calls.
• Example (feature engineering):

import numpy as np

# Acceleration magnitude
df_resampled['acc_mag'] = np.sqrt(df_resampled['ax_clean']**2 +
                                 df_resampled['ay_clean']**2 +
                                 df_resampled['az_clean']**2)

# Simple activity label
conditions = [
    (df_resampled['acc_mag'] < 0.1),
    (df_resampled['acc_mag'] >= 0.1) & (df_resampled['acc_mag'] <= 1.5)
]
choices = ['rest', 'walk']
df_resampled['activity'] = np.select(conditions, choices, default='run')

5. Data Quality Assurance

• Outlier detection: Use z‑score or IQR methods to flag anomalous readings.
• Consistency checks: Verify that GPS points are within plausible distance ranges (e.g., not jumping > 50 km in a minute).
• Logging & audit trail: Record preprocessing steps, parameter values, and any data discarded for reproducibility.

6. Storage & Export

• Save the cleaned dataset as Parquet or Feather for fast read/write in downstream analysis.
• Maintain a separate “raw” folder to preserve original sensor dumps for future reprocessing.
• Example:

df_resampled.to_parquet('cleaned_canine_data.parquet')

Practical Tips & Common Pitfalls

• Sync clocks across devices: Even a few seconds drift can misalign multi‑modal data.
• Beware of over‑filtering: Aggressive smoothing may erase subtle behavioral cues (e.g., short bursts of excitement).
• Document assumptions: Clearly state why you chose a particular imputation method or filter cutoff—this aids peer review and future replication.
• Automate pipelines: Use tools like Airflow, Prefect, or simple cron jobs to run preprocessing nightly.

By rigorously cleaning and enriching your raw sensor data, you lay a solid foundation for uncovering reliable canine behavior trends. The processed dataset becomes a trustworthy asset that can drive insights—from individual pet care recommendations to large‑scale studies on how dogs interact with their environments.

Feature Extraction from Movement Sensors

# Load raw sensor data
import pandas as pd
df = pd.read_csv('dog_sensor_data.csv', parse_dates=['timestamp'])
# Resample to uniform 50 Hz if necessary
df_resampled = df.set_index('timestamp').resample('20ms').mean().interpolate()

# Sliding window (5 s, 50% overlap)
window_size = 250   # samples at 50 Hz
step = 125

features_list = []
for start in range(0, len(df_resampled)-window_size+1, step):
    win = df_resampled.iloc[start:start+window_size]
    feat = {}
    
    # Time‑domain
    for axis in ['acc_x', 'acc_y', 'acc_z']:
        feat[f'{axis}_mean'] = win[axis].mean()
        feat[f'{axis}_std']  = win[axis].std()
        feat[f'{axis}_skew']= win[axis].skew()
    
    # Magnitude
    mag = (win[['acc_x','acc_y','acc_z']]**2).sum(axis=1)**0.5
    feat['mag_mean'] = mag.mean()
    feat['sma']      = mag.sum() / window_size
    
    # Frequency‑domain
    fft_vals = np.abs(np.fft.rfft(mag))
    freqs   = np.fft.rfftfreq(len(mag), d=0.02)  # 50 Hz sampling
    feat['dom_freq'] = freqs[np.argmax(fft_vals)]
    
    features_list.append(feat)

features_df = pd.DataFrame(features_list)

Heart Rate Variability and Stress Indicators

Heart rate variability (HRV) is a non‑invasive, objective measure of the autonomic nervous system’s regulation of cardiac activity. In dogs, higher HRV typically reflects greater parasympathetic tone (relaxation), while lower HRV indicates sympathetic dominance (stress or arousal). When combined with behavioral observations and environmental context, HRV can reveal subtle shifts in a dog’s emotional state that are otherwise difficult to detect.

Key HRV Metrics for Canine Studies

• RMSSD (Root Mean Square of Successive Differences): Sensitive to parasympathetic activity; calculated from beat‑to‑beat intervals.
• SDNN (Standard Deviation of NN Intervals): Overall HRV, reflects both sympathetic and parasympathetic contributions.
• LF/HF Ratio: Low‑frequency to high‑frequency power ratio; higher values suggest sympathetic dominance.
• NN50 / pNN50: Count of successive RR interval differences > 50 ms, expressed as a percentage.

Collecting Sensor Data in the Field

Modern canine collars (e.g., Garmin, Whistle, or custom Arduino‑based units) can capture ECG or photoplethysmography (PPG) data at sampling rates of 50–200 Hz. To ensure high‑quality HRV calculations:

1. Use a collar with continuous recording and low motion artifact filtering.
2. Synchronize timestamps across devices (GPS, accelerometer) to align physiological data with behavioral events.
3. Store raw RR intervals in a CSV or SQLite database for post‑processing.

Practical Analysis Workflow

# Python example: Compute RMSSD from RR interval file
import pandas as pd
import numpy as np

# Load RR intervals (ms)
rr = pd.read_csv('rr_intervals.csv')['RR'].values

# Convert to seconds for consistency
rr_sec = rr / 1000.0

# Successive differences
diffs = np.diff(rr_sec)

# RMSSD calculation
rmssd = np.sqrt(np.mean(diffs ** 2))
print(f'RMSSD: {rmssd:.3f} s')

After computing HRV metrics, merge them with behavioral logs (e.g., timestamps of play, training, feeding). A simple example using pandas:

# Merge HRV and behavior data
hrv = pd.read_csv('hrv_metrics.csv', parse_dates=['timestamp'])
behavior = pd.read_csv('behavior_log.csv', parse_dates=['event_time'])

# Align by nearest timestamp within 5‑second window
merged = pd.merge_asof(hrv.sort_values('timestamp'),
                      behavior.sort_values('event_time'),
                      left_on='timestamp',
                      right_on='event_time',
                      direction='nearest',
                      tolerance=pd.Timedelta('5s'))

print(merged.head())

Interpreting Stress Indicators in Context

• Baseline HRV: Establish a resting baseline for each dog (e.g., during quiet sleep). Deviations > 20% from baseline may signal acute stress.
• Event‑Triggered Changes: Sudden drops in RMSSD coinciding with a new environment or loud noise suggest sympathetic activation.
• Longitudinal Trends: Gradual decline in HRV over weeks can indicate chronic anxiety or health issues (e.g., pain, cardiac disease).

Case Study: Training Session Analysis

A working‑dog team used collar‑based ECG to monitor HRV during obedience training. By plotting RMSSD against the trainer’s commands, they observed:

1. “Sit” command: Stable RMSSD (~0.45 s), indicating calm compliance.
2. “Recall” in a noisy park: RMSSD dropped to ~0.25 s, correlating with increased heart rate and behavioral agitation.

Tips for Practitioners

• Always pair physiological data with high‑resolution activity logs (accelerometer, GPS).
• Use a rolling average window (e.g., 5 min) to smooth HRV signals before event alignment.
• Consider individual variability: breed, age, and prior training can influence baseline HRV ranges.
• Integrate machine‑learning classifiers (SVM, Random Forest) trained on labeled stress vs. relaxed states for automated real‑time alerts.

Resources & Further Reading

• HRV in Veterinary Medicine: A Review
• Open‑source dog HRV analysis toolkit
• Manufacturer data sheets for collar ECG modules (Garmin, Whistle, etc.)

Environmental Contextualization (Temperature, Noise)

When studying canine behavior through wearable or environmental sensors, it is essential to contextualize the data with ambient conditions such as temperature and noise levels. These factors can profoundly influence a dog’s activity patterns, stress responses, and overall well‑being.

Why Temperature Matters

• Thermoregulation: Dogs rely on panting, increased heart rate, and behavioral changes (e.g., seeking shade) to regulate body temperature. Sensor data that captures core or skin temperature can help identify heat‑stress episodes.
• Activity Modulation: Higher ambient temperatures often reduce activity levels, especially during peak sun hours. Conversely, cooler mornings may see increased play and exploration.
• Breed & Coat Differences: Thick‑coated breeds (e.g., Siberian Husky) exhibit different temperature thresholds compared to short‑haired breeds (e.g., Jack Russell Terrier).

Noise Levels as a Stressor

• Auditory Sensitivity: Dogs have a wider hearing range than humans. Sudden loud noises (traffic, fireworks) can trigger startle responses, increased heart rate, and vocalization.
• Behavioral Indicators: Accelerometer data may show rapid micro‑movements or attempts to escape when noise spikes are detected by an integrated microphone.
• Long‑Term Impact: Chronic exposure to high ambient noise can lead to anxiety, reduced sleep quality, and altered social interactions.

Integrating Sensor Data with Environmental Metrics

Below is a step‑by‑step guide on how to combine temperature and noise data with behavioral metrics from canine wearables:

1. Data Collection: Use a multi‑sensor collar that records GPS, accelerometry, skin temperature, and ambient sound levels at 1 Hz.
2. Time‑Syncing: Align all data streams using timestamps. Convert UTC to local time for better interpretation of daily patterns.
3. Feature Extraction:
   • Compute rolling averages (e.g., 5 min) for temperature and decibel levels.
   • Identify peaks in sound intensity (>70 dB) and correlate with sudden changes in accelerometer counts.
   • Calculate activity bouts: periods where acceleration exceeds a threshold (e.g., >1.2 g).
4. Statistical Analysis:
   • Run regression models to test the relationship between temperature and activity level.
   • Use mixed‑effects models to account for individual dog variability.
   • Apply time‑series decomposition (trend, seasonality) to separate environmental effects from intrinsic behavior patterns.
5. Visualization:
   • Create dual‑axis line charts: temperature on the primary y‑axis and activity counts on the secondary y‑axis.
   • Overlay noise spikes as vertical bars to show temporal alignment with behavior changes.

Practical Advice for Researchers & Pet Owners

• Choose the Right Sensors: Ensure microphones have a flat frequency response up to 20 kHz and temperature probes are calibrated for skin contact.
• Set Thresholds Early: Define what constitutes “high” temperature or noise based on breed-specific tolerances before data collection begins.
• Plan Data Retention: Store raw sensor streams in a secure database with automated backups. Use cloud storage to facilitate collaboration across research sites.
• Consider Ethical Implications: Inform owners about potential stressors and provide guidelines on mitigating heat or noise exposure (e.g., using cooling vests, sound‑proofing rooms).
• Validate Findings with Observational Studies: Complement sensor data with video recordings to confirm behavioral interpretations.

Case Study: Heat Stress in a Working Border Collie

A Border Collie working in a temperate climate wore a collar that logged skin temperature, heart rate, and GPS. During a 90‑minute field test on a hot day (ambient 32 °C), the dog’s skin temperature rose from 36.5 °C to 39.2 °C. Accelerometer data showed a marked drop in activity after the first 30 minutes, coinciding with increased heart rate and a spike in ambient noise (traffic at 75 dB). The owner adjusted the training schedule to cooler hours, resulting in a 25% increase in daily activity over the following week.

By contextualizing sensor data with temperature and noise metrics, researchers and owners can identify environmental triggers of stress or reduced performance, leading to proactive interventions that enhance canine welfare and productivity.

Behavior Classification Algorithms

Temporal Patterns in Daily Activity

Anomaly Detection and Health Alerts

Integrating Owner Feedback with Sensor Insights

While wearable sensors deliver objective, high‑frequency data on a dog’s activity and physiological state, they can only tell part of the story. Owners are the first line of observation, and their subjective insights—what a dog seems to enjoy or dislike, how it reacts in specific situations—provide essential context that raw numbers cannot capture. By combining these two streams, you get a richer, more actionable picture of canine behavior.

Why Owner Input Matters

• Contextual nuance: A sudden spike in heart rate during a walk could be due to an exciting scent, a stressful encounter with another dog, or simply the thrill of running. Owners can often identify which scenario occurred.
• Behavioral cues: Subtle signs—ear position, tail wag speed, whining—are difficult for algorithms to interpret without human annotation.
• Long‑term trends: Owners notice changes over months or seasons that a short‑term sensor study might miss (e.g., gradual anxiety during thunderstorms).

Practical Ways to Capture Owner Feedback

1. Daily log sheets: A simple spreadsheet with columns for date, time, activity, mood rating (1–5), and notes. Encourage owners to record immediately after the event.
2. Mobile app prompts: Push notifications that pop up at key times—after a walk, after a vet visit, or when sensor data shows unusual patterns—to prompt quick feedback.
3. Photo/video diaries: Owners can attach images or short clips of the dog during specific moments. These visual records help researchers match sensor spikes to observable events.
4. Structured questionnaires: Periodic surveys (e.g., monthly) asking about perceived changes in energy, appetite, or anxiety levels.

Techniques for Merging Data Streams

• Time‑syncing: Ensure all logs use the same timestamp format. Sensors typically record UTC; convert owner entries to UTC before merging.
• Event tagging: Assign tags (e.g., “walk,” “vet visit,” “playtime”) to both sensor data segments and owner notes. This facilitates automated cross‑matching.
• Annotation tools: Use platforms like LabVantage or custom dashboards where owners can annotate sensor graphs directly, marking peaks with descriptive labels.
• Machine learning enrichment: Train classifiers on labeled data (owner‑annotated segments) to predict behavioral states from sensor features alone in future studies.

Case Study: Tracking Anxiety During Storms

A research team monitored a group of dogs with heart rate monitors and accelerometers over six months. Owners recorded weather conditions and noted any signs of distress. By overlaying the data, researchers discovered that:

• Heart rate increased 30% during thunderstorms.
• Accelerometer patterns showed reduced movement but frequent pacing.
• Owner logs consistently reported whining and panting.

The combined analysis led to a targeted intervention—calming music played during storms—which reduced heart rate spikes by 18%. This example illustrates how owner feedback can validate sensor findings and guide practical solutions.

Tips for Maximizing Data Quality

1. Educate owners: Provide short tutorials on interpreting basic sensor outputs (e.g., what a spike means) so they can give more precise feedback.
2. Simplify logging: Use check‑boxes or sliders rather than free text where possible to reduce effort and increase consistency.
3. Incentivize participation: Offer small rewards (e.g., discounts on pet products) for regular log completion.
4. Protect privacy: Anonymize data before sharing with researchers; obtain explicit consent for any photos or videos used in analysis.

By weaving owner observations into the fabric of sensor‑based studies, you unlock a more holistic understanding of canine behavior trends—an essential step toward improving welfare and tailoring interventions to each dog’s unique needs.

Case Studies: Training, Rehabilitation, and Socialization

Conclusion

FAQ

// Example: Log accelerometer data
#include <Wire.h>
#include "Adafruit_MPU6050.h"

Adafruit_MPU6050 mpu;

void setup() {
  Serial.begin(115200);
  if (!mpu.begin()) { Serial.println("Failed to find MPU6050"); while (1); }
}

void loop() {
  sensors_event_t a, g, temp;
  mpu.getEvent(&a, &g, &temp);
  Serial.print(a.acceleration.x); Serial.print(",");
  Serial.print(a.acceleration.y); Serial.print(",");
  Serial.println(a.acceleration.z);
  delay(1000);
}