AutopilotSim2驾驶模拟器使用

阳光穿透心脏的1/2处 2022-12-01 05:09 87阅读 0赞

1. 运行

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzE1MjA0MTc5_size_16_color_FFFFFF_t_70][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzE1MjA0MTc5_size_16_color_FFFFFF_t_70 1][]

卷积训练,加载收集的数据

import numpy as np 
    import pandas as pd
    data = pd.read_csv('info.csv')
    from sklearn.model_selection import train_test_split
    ##数据拆分 训练集与测试集
    X_train,X_test,y_train,y_test = train_test_split(data['center'],data['steering'],test_size=0.2,random_state=25)
    print("训练集：{}个数据，测试集：{}个数据".format(len(X_train),len(X_test)))
    
    from imgaug import augmenters as iaa
    import numpy as np 
    import cv2
    import matplotlib.pyplot as plt
    def zoom_img(img):
        zoom = iaa.Affine(scale=(1.0,1.3))
        return zoom.augment_image(img)
    
    def trans_img(img):
        trans = iaa.Affine(translate_percent={'x':(-0.1,0.1),'y':(-0.1,0.1)})
        return trans.augment_image(img)
    def img_random_brightness(img):
        brightness = iaa.Multiply((0.3,1.2))
        return brightness.augment_image(img)
    
    def img_flip(img,steering_angle):
        img = cv2.flip(img,1)
        steering_angle  = -steering_angle
        return img,steering_angle
    
    ## 自定义的算法组合，数据增强器,path  图片的路径,steering_angle 方向盘的角度
    def random_augment(path,steering_angle):
        image = plt.imread(path)
        if np.random.rand()<0.5:
            image = zoom_img(image)
        if np.random.rand()<0.5:
            image = trans_img(image)
        if np.random.rand()<0.5:
            image = img_random_brightness(image)
        if np.random.rand()<0.5:
            image,steering_angle = img_flip(image,steering_angle)
        
        return image,steering_angle
    
    ## 训练集数据增强器
    def train_batch_generator(batch_size):
        while True:
            batch_img = []
            batch_angle =[]
            for i in range(batch_size):
                index = np.random.randint(0,len(X_train))
                img , angle = random_augment(X_train.iloc[index],y_train.iloc[index])
                img = img[60:135,:,:]
                img = cv2.GaussianBlur(img,(3,3),1)
                img = cv2.resize(img,(200,66))
                img = cv2.cvtColor(img,cv2.COLOR_RGB2YUV)
                batch_img.append(img/255.0)
                batch_angle.append(angle)
            yield np.array(batch_img), np.array(batch_angle)
    
    ## 测试集数据增强器
    def test_batch_generator(batch_size):
        while True:
            batch_img = []
            batch_angle =[]
            for i in range(batch_size):
                index = np.random.randint(0,len(X_test))
                img , angle = random_augment(X_test.iloc[index],y_test.iloc[index])
                img = img[60:135,:,:]
                img = cv2.GaussianBlur(img,(3,3),1)
                img = cv2.resize(img,(200,66))
                img = cv2.cvtColor(img,cv2.COLOR_RGB2YUV)
                batch_img.append(img/255.0)
                batch_angle.append(angle)
            yield np.array(batch_img), np.array(batch_angle)
    
    ## 提升模型的泛化能力，增强稳定度
    # 回归问题 --->(-1,1)任意的一个值  mse误差函数
    import tensorflow as tf
    model = tf.keras.Sequential()
    model.add(tf.keras.layers.Conv2D(24,(5,5),strides=(2,2),input_shape=(66,200,3),activation='elu'))
    
    model.add(tf.keras.layers.Dropout(0.5))
    
    model.add(tf.keras.layers.Conv2D(36,(5,5),strides=(2,2),activation='elu'))
    
    model.add(tf.keras.layers.Conv2D(48,(5,5),strides=(2,2),activation='elu'))
    
    model.add(tf.keras.layers.Dropout(0.5))
    
    model.add(tf.keras.layers.Conv2D(64,(3,3),activation='elu'))
    
    model.add(tf.keras.layers.Conv2D(64,(3,3),activation='elu'))
    
    model.add(tf.keras.layers.Dropout(0.5))
    
    model.add(tf.keras.layers.Flatten())
    
    model.add(tf.keras.layers.Dense(100,activation='elu'))
    
    model.add(tf.keras.layers.Dropout(0.5))
    
    model.add(tf.keras.layers.Dense(50,activation='elu'))
    
    model.add(tf.keras.layers.Dense(10,activation='elu'))
    
    model.add(tf.keras.layers.Dropout(0.2))
    
    model.add(tf.keras.layers.Dense(1)) ## 方向盘的角度
    
    model.summary()
    
    model.compile( optimizer='rmsprop',loss='mse')
    from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau
    
    ## 预防训练时意外中断训练
    checkpoint = ModelCheckpoint("./self_driving_model.h5",
                                 monitor="val_loss",
                                 mode="min",
                                 save_best_only = True,
                                 verbose=1)
    #earlystopping 训练到第n的时候停下来,最优的时候自己停下来 
    earlystop = EarlyStopping(monitor = 'val_loss', 
                              min_delta = 0, 
                              patience = 3,
                              verbose = 1,
                              restore_best_weights = True)
    ##动态调整学习速率
    reduce_lr = ReduceLROnPlateau(monitor = 'val_loss', factor = 0.2, patience = 3, verbose = 1, min_delta = 0.0001)
    
    # we put our call backs into a callback list
    callbacks = [earlystop, checkpoint, reduce_lr]
    ##训练
    history = model.fit_generator(train_batch_generator(1000),
                                  steps_per_epoch = 30,
                                  epochs =10,
                                  validation_data = test_batch_generator(300),
                                  validation_steps = 10  ,
                                  callbacks = callbacks
                                 )
    ##保存模型
    model.save("001.h5")
    ##读取模型
    save_model = tf.keras.models.load_model('001.h5')
    
    ## 随机选一张测试集的照片
    index_test = np.random.randint(len(X_test_imgs))
    index_test
    img = plt.imread(X_test.iloc[index_test])
    plt.imshow(img)
    save_model.predict(X_test_imgs[index_test].reshape(1,66,200,3))

车辆自动驾驶代码,打开AutopilotSim2自动驾驶模拟器自动驾驶模式,加载自己的模型运行以下代码车就可以跑了

#安装python开发web程序的依赖
    #!conda install flask
    #!pip install imgaug
    #!conda install -c conda-forge python-socketio
    #!conda install -c conda-forge eventlet
    ## 包装方法
    ## 图片预处理的逻辑，接受参数图片的路径
    def img_preprocess(src_image):
       
        img = src_image[60:135,:,:]
        img = cv2.GaussianBlur(img,(3,3),1)
        img = cv2.resize(img,(200,66))
        img = cv2.cvtColor(img,cv2.COLOR_RGB2YUV)
        return img
    
    import tensorflow as tf
    #save_model = tf.keras.models.load_model('hahaha.h5')
    save_model = tf.keras.models.load_model('003.h5')
    ## 从模拟器汽车的摄像头里面读到数据， 基于图像输入，使用cnn训练好的模型，预测方向盘的角度。
    
    import eventlet
    import socketio
    from flask import Flask
    from io import BytesIO
    from PIL import Image
    import cv2
    import base64
    import numpy as np
    
    sio = socketio.Server()  ## 创建socketio的服务器
    app = Flask(__name__)
    
    speed_limit = 30
    
    @sio.on('connect')
    def connect(sid,env):
        print("websocket链接建立了")
    #     sio.emit("steer",data={
    #         'steering_angle':'0',
    #         'throttle':'0.0'
    #     })
    
    
    def send_control(angle,throttle):
        sio.emit("steer",data={
            'steering_angle':angle.__str__(),
            'throttle':throttle.__str__()
        })
    
        
    @sio.on('telemetry')
    def telemetry(sid,data):
        image = Image.open(BytesIO(base64.b64decode(data['image'])))
        image = np.asarray(image)
        cv2.imshow("info",image)
        cv2.waitKey(10)
        
        #预测
        image = img_preprocess(image)
        image = image / 255.0
        angel = save_model.predict(image.reshape(1,66,200,3))[0][0]
        
        ##当前的车速
        speed = float(data['speed'])
        throttle = 1.0 - speed / speed_limit
        
        send_control(angel,throttle)
        
    if __name__=='__main__':
        # 把web服务器和socketio绑定在一起。
        app = socketio.Middleware(sio,app)
        ## 监听socket的事件了
        eventlet.wsgi.server(eventlet.listen(('',4567)),app)

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzE1MjA0MTc5_size_16_color_FFFFFF_t_70]: /images/20221123/b1bb39ae61134aea8852c706fcb318ac.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzE1MjA0MTc5_size_16_color_FFFFFF_t_70 1]: /images/20221123/42791a873d27449ebe2c17791d95d1e6.png