Android Open Source - IntelligentCarForAndroid Predictor

• Back to Network/car  ↑

From Project

Back to project page IntelligentCarForAndroid.

License

The source code is released under:

Apache License

Java Source Code

package org.davidsingleton.core;
/*from  www.  ja  v a2  s . c om*/
import java.util.Queue;
import java.util.concurrent.ConcurrentLinkedQueue;


/**
 * A class which makes neural network predictions from the most recent set of features added to its
 * (concurrent) queue.  Should be run in its own Thread.
 */
public class Predictor implements Runnable {
  
  private static final double NN_CONFIDENCE_THRESHOLD = 0.4;

  public interface PredictionListener {
    void onPrediction(double[] pred, boolean left, boolean right, boolean forward, boolean reverse);
  }

  private Queue<byte[]> predictQueue = new ConcurrentLinkedQueue<byte[]>();
  
  private Thread predictThread;
  private PredictionListener listener;
  private NeuralNetwork nn;

  public Predictor(PredictionListener listener, NeuralNetwork nn) {
    this.listener = listener;
    this.nn = nn;
    predictThread = new Thread(this);
    predictThread.start();
  }

  @Override
  public void run() {
    while (true) {
      if (!predictQueue.isEmpty()) {
        byte[] features = null;

        synchronized (predictQueue) {
          int dropped = -1;
          while (!predictQueue.isEmpty()) {
            // Take the freshest features in the queue, dropping the rest
            features = predictQueue.remove();
            dropped = dropped + 1;
          }
          if (dropped > 0) {
            System.out.println(dropped + " frames dropped from predict queue");
          }
        }
        predict(features);
      } else {
        try {
          Thread.sleep(10);
        } catch (InterruptedException e) {
        }
      }
    }
  }

  public void queuePredict(byte[] features) {
    synchronized (predictQueue) {
      predictQueue.add(features);
    }
  }
  
  private void predict(byte[] features) {
    double[] pred = nn.predict(features);

    boolean left = pred[0] > NN_CONFIDENCE_THRESHOLD;
    boolean right = pred[1] > NN_CONFIDENCE_THRESHOLD;
    boolean forward = pred[2] > NN_CONFIDENCE_THRESHOLD;
    boolean reverse = pred[3] > NN_CONFIDENCE_THRESHOLD;

    listener.onPrediction(pred, left, right, forward, reverse);
  }
}

Java Source Code List

• Back to Network/car  ↑