Tensorflow based Object Detection Using Raspberry pi - Eye of Robots

Object Detection using Raspberry Pi:

Why?

• Object detection is a kind of eye for AI or Machines of Future. Just Imagine a Humanoid Robot can not visualize like human hence it needs to detect what it is watching or where it is?

• Detection is always a curious case, Deep Learning covers a major area on this.

• Here we use a camera, to be precise pixels to find the object and also classify them.4

How?

• Installation of TensorFlow in raspberry pi,

               This can be done in two ways, either using Docker and compile tensorflow inside it or installing with a compiled file.

Here we go for Compiled file from github repository,

Open terminal and make directory tf

mkdir tf

cd tf

Download the wheel file using wget

wget https://github.com/lhelontra/tensorflow-on-arm/releases/download/v1.13.1/tensorflow-1.13.1-cp35-none-linux_armv7l.whl

This downloads the tensorflow file and ready to install

sudo pip3 install tensorflow-1.13.1-cp35-none-linux_armv7l.whl

This is tensorflow 1.13 version for raspberry pi, if you see any update find here, 

https://github.com/lhelontra/tensorflow-on-arm/releases

Tensorflow needs additional package of LibAtlas,lets install that too

sudo apt-get install libatlas-base-dev

Camera object detection is made through openCV hence we need other dependencies with respect to tensorflow and OpenCV,

sudo pip3 install pillow lxml jupyter matplotlib cython

sudo apt-get install python-tk

Sometimes you may get error on this installation, if you get error then try installing one by one like,

sudo pip3 install matplotlib

sudo pip3 install cpython

• OpenCV installation

Image plotting is done through matplotlib and OpenCV is core in image processing and tensorflow uses pixel analysis for producing the results.

Hence, OpenCV installation is must one,For installing it in raspberry pi there are few dependecies,

sudo apt-get install libjpeg-dev libtiff5-dev libjasper-dev libpng12-dev

sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev

sudo apt-get install libxvidcore-dev libx264-dev

sudo apt-get install qt4-dev-tools

Now, lets start OpenCV install command,

sudo pip3 install opencv-python

if you really interested in learning more on OpenCV, find pyimagesearch.com provides valuable information.

• Protobuf installation

 Protobuf, a package that implements Google’s Protocol Buffer data format is used by Tensorflow for object detection.

First, get the packages needed to compile Protobuf from source. 

sudo apt-get install autoconf automake libtool curl

Then download the protobuf release from its GitHub repository by issuing:

wget https://github.com/google/protobuf/releases/download/v3.5.1/protobuf-all-3.5.1.tar.gz

Unpack the file and cd into the folder:

tar -zxvf protobuf-all-3.5.1.tar.gz

cd protobuf-3.5.1

Configure the build by issuing the following command (it takes about 2 minutes):

./configure

Build the package

make

The build process took 1 hr 10 minutes on my Raspberry Pi. When it’s finished, 

make check 

This process takes even longer, clocking in at 1.5 hrs on my Pi. If you see errors, you can ignore them for now. Now that it’s built, install it by issuing:

sudo make install

Then move into the python directory and export the library path:

cd python

export LD_LIBRARY_PATH=../src/.libs

Next, 

python3 setup.py build --cpp_implementation 

python3 setup.py test --cpp_implementation

sudo python3 setup.py install --cpp_implementation

Then enter the following path commands:

export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=cpp

export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION_VERSION=3

Finally, finish with the command

sudo ldconfig

Now Protobuf is installed on the Pi. Verify it’s installed correctly by issuing the command below and making sure it puts out the default help text.

protoc

then, Raspberry Pi needs to be restarted after this process, or TensorFlow will not work. 

sudo reboot now

• Tensorflow Config

Now, back to tensorflow,

let start with directory structure,

mkdir tensorflow

cd tensorflow

Get the tensorflow repostiory ready by downloading it with wget

git clone --recurse-submodules https://github.com/tensorflow/models.git

Since we didn't have docker here, we need to configure path for python to enable tensorflow working structure,

for which edit the following file,

sudo nano ~/.bashrc

In the file, enter the last line and add the following text,

export PYTHONPATH=$PYTHONPATH:/home/pi/tensorflow/models/research:/home/pi/tensorflow/models/research/slim

press ctrl+x and save it by providing 'yes'

We need to use Protoc to compile the Protocol Buffer (.proto) files used by the Object Detection API. The .proto files are located in /research/object_detection/protos, but we need to execute the command from the /research directory. Issue:

cd /home/pi/tensorflow/models/research

protoc object_detection/protos/*.proto --python_out=.

Now, move into object detection folder,

cd /home/pi/tensorflow/models/research/object_detection

Download the Tensorflow detection model with wget,

wget http://download.tensorflow.org/models/object_detection/ssdlite_mobilenet_v2_coco_2018_05_09.tar.gz

extract the file,

tar -xzvf ssdlite_mobilenet_v2_coco_2018_05_09.tar.g

Here comes the end of sequencial order,

Final Step of Running Detection model,

Enter into the directory, 

cd /home/pi/tensorflow/models/research/object_detection

Create a empty file named, objectdetection.py 

Copy the below content into the file and run from terminal using python3 objectdetection.py
######## Picamera Object Detection Using Tensorflow Classifier #########


# This program uses a TensorFlow classifier to perform object detection.
# It loads the classifier uses it to perform object detection on a Picamera feed.
# It draws boxes and scores around the objects of interest in each frame from
# the Picamera. It also can be used with a webcam by adding "--usbcam"
# when executing this script from the terminal.

## Some of the code is copied from Google's example at
## https://github.com/tensorflow/models/blob/master/research/object_detection/object_detection_tutorial.ipynb

## and some is copied from Dat Tran's example at
## https://github.com/datitran/object_detector_app/blob/master/object_detection_app.py

## but I changed it to make it more understandable to me.


# Import packages
import os
import cv2
import numpy as np
from picamera.array import PiRGBArray
from picamera import PiCamera
import tensorflow as tf
import argparse
import sys

# Set up camera constants
IM_WIDTH = 1280
IM_HEIGHT = 720
#IM_WIDTH = 640    Use smaller resolution for
#IM_HEIGHT = 480   slightly faster framerate

# Select camera type (if user enters --usbcam when calling this script,
# a USB webcam will be used)
camera_type = 'picamera'
parser = argparse.ArgumentParser()
parser.add_argument('--usbcam', help='Use a USB webcam instead of picamera',
                    action='store_true')
args = parser.parse_args()
if args.usbcam:
    camera_type = 'usb'

# This is needed since the working directory is the object_detection folder.
sys.path.append('..')

# Import utilites
from utils import label_map_util
from utils import visualization_utils as vis_util

# Name of the directory containing the object detection module we're using
MODEL_NAME = 'ssdlite_mobilenet_v2_coco_2018_05_09'

# Grab path to current working directory
CWD_PATH = os.getcwd()

# Path to frozen detection graph .pb file, which contains the model that is used
# for object detection.
PATH_TO_CKPT = os.path.join(CWD_PATH,MODEL_NAME,'frozen_inference_graph.pb')

# Path to label map file
PATH_TO_LABELS = os.path.join(CWD_PATH,'data','mscoco_label_map.pbtxt')

# Number of classes the object detector can identify
NUM_CLASSES = 90

## Load the label map.
# Label maps map indices to category names, so that when the convolution
# network predicts `5`, we know that this corresponds to `airplane`.
# Here we use internal utility functions, but anything that returns a
# dictionary mapping integers to appropriate string labels would be fine
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories)

# Load the Tensorflow model into memory.
detection_graph = tf.Graph()
with detection_graph.as_default():
    od_graph_def = tf.GraphDef()
    with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
        serialized_graph = fid.read()
        od_graph_def.ParseFromString(serialized_graph)
        tf.import_graph_def(od_graph_def, name='')

    sess = tf.Session(graph=detection_graph)


# Define input and output tensors (i.e. data) for the object detection classifier

# Input tensor is the image
image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')

# Output tensors are the detection boxes, scores, and classes
# Each box represents a part of the image where a particular object was detected
detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')

# Each score represents level of confidence for each of the objects.
# The score is shown on the result image, together with the class label.
detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')

# Number of objects detected
num_detections = detection_graph.get_tensor_by_name('num_detections:0')

# Initialize frame rate calculation
frame_rate_calc = 1
freq = cv2.getTickFrequency()
font = cv2.FONT_HERSHEY_SIMPLEX

# Initialize camera and perform object detection.
# The camera has to be set up and used differently depending on if it's a
# Picamera or USB webcam.

# I know this is ugly, but I basically copy+pasted the code for the object
# detection loop twice, and made one work for Picamera and the other work
# for USB.

### Picamera ###
if camera_type == 'picamera':
    # Initialize Picamera and grab reference to the raw capture
    camera = PiCamera()
    camera.resolution = (IM_WIDTH,IM_HEIGHT)
    camera.framerate = 10
    rawCapture = PiRGBArray(camera, size=(IM_WIDTH,IM_HEIGHT))
    rawCapture.truncate(0)

    for frame1 in camera.capture_continuous(rawCapture, format="bgr",use_video_port=True):

        t1 = cv2.getTickCount()
        
        # Acquire frame and expand frame dimensions to have shape: [1, None, None, 3]
        # i.e. a single-column array, where each item in the column has the pixel RGB value
        frame = np.copy(frame1.array)
        frame.setflags(write=1)
        frame_expanded = np.expand_dims(frame, axis=0)

        # Perform the actual detection by running the model with the image as input
        (boxes, scores, classes, num) = sess.run(
            [detection_boxes, detection_scores, detection_classes, num_detections],
            feed_dict={image_tensor: frame_expanded})

        # Draw the results of the detection (aka 'visulaize the results')
        vis_util.visualize_boxes_and_labels_on_image_array(
            frame,
            np.squeeze(boxes),
            np.squeeze(classes).astype(np.int32),
            np.squeeze(scores),
            category_index,
            use_normalized_coordinates=True,
            line_thickness=8,
            min_score_thresh=0.40)

        cv2.putText(frame,"FPS: {0:.2f}".format(frame_rate_calc),(30,50),font,1,(255,255,0),2,cv2.LINE_AA)

        # All the results have been drawn on the frame, so it's time to display it.
        cv2.imshow('Object detector', frame)

        t2 = cv2.getTickCount()
        time1 = (t2-t1)/freq
        frame_rate_calc = 1/time1

        # Press 'q' to quit
        if cv2.waitKey(1) == ord('q'):
            break

        rawCapture.truncate(0)

    camera.close()

### USB webcam ###
elif camera_type == 'usb':
    # Initialize USB webcam feed
    camera = cv2.VideoCapture(0)
    ret = camera.set(3,IM_WIDTH)
    ret = camera.set(4,IM_HEIGHT)

    while(True):

        t1 = cv2.getTickCount()

        # Acquire frame and expand frame dimensions to have shape: [1, None, None, 3]
        # i.e. a single-column array, where each item in the column has the pixel RGB value
        ret, frame = camera.read()
        frame_expanded = np.expand_dims(frame, axis=0)

        # Perform the actual detection by running the model with the image as input
        (boxes, scores, classes, num) = sess.run(
            [detection_boxes, detection_scores, detection_classes, num_detections],
            feed_dict={image_tensor: frame_expanded})

        # Draw the results of the detection (aka 'visulaize the results')
        vis_util.visualize_boxes_and_labels_on_image_array(
            frame,
            np.squeeze(boxes),
            np.squeeze(classes).astype(np.int32),
            np.squeeze(scores),
            category_index,
            use_normalized_coordinates=True,
            line_thickness=8,
            min_score_thresh=0.85)

        cv2.putText(frame,"FPS: {0:.2f}".format(frame_rate_calc),(30,50),font,1,(255,255,0),2,cv2.LINE_AA)
        
        # All the results have been drawn on the frame, so it's time to display it.
        cv2.imshow('Object detector', frame)

        t2 = cv2.getTickCount()
        time1 = (t2-t1)/freq
        frame_rate_calc = 1/time1

        # Press 'q' to quit
        if cv2.waitKey(1) == ord('q'):
            break

    camera.release()

cv2.destroyAllWindows()