Object-detection-raspberry-pi

install tensorflow and Object detection on raspberry pi

install tensorflow

TensorFlow：1.9
Python：3.5.3
Hardware: Raspberry 3B
install ibatlas-base-dev

sudo apt-get install libatlas-base-dev
Do you want to continue? [Y/n] input：Y

install tensorflow

sudo pip3 install  tensorflow

install other Dependent tools :numpy，tensorboard，markdown .....
in my case is tensorflow 1.9 so numpy version is 1.13.0,if numpy 1.16 or higher is not work

install Object detetion model

see https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md

install opencv

see https://www.learnopencv.com/install-opencv-4-on-raspberry-pi/

Ubuntu VNC

sudo apt-get update
sudo apt-get upgrade
sudo apt-get install tightvncserver
vncserver -geometry 800×800 :1

open Remote Desktop Viewer and Connect with you IP

Collecting Datasets via Raspberry Pi Camera

if use raspberrt pi camera: Because my video is reddish, so i change the white balance

from picamera import PiCamera
camera = PiCamera()
camera.awb_mode='off'
#change white balance 
camera.awb_gains=(1.0,1.2)
camera.resolution =(640 ,480)

if use usb camera:

import cv2
video = cv2.VideoCapture(0)

Use opencv Saving Video Frames

cap = cv2.VideoCapture("your video name")

Lable the ground truth

see https://github.com/tzutalin/labelImg save as xml file

XML file to CSV

see file XML file to CSV

CSV to TFCODE

see file CSV to TFCODE

Create a .pbtxt file/label map file

Create a train.pbtxt file in object_detection/data

item {
 id: 1
 name: 'car'
}

Change .configfile

Download model https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md
in my case ssd_mobilenet_v1_coco.config Items to modify in the .config:

• model.ssd.num_classes
  
• train_config.fine_tune_checkpoint
  
• train_input_reader.label_map_path
  
• train_input_reader.tf_record_input_reader
  
• eval_input_reader.label_map_path
  
• eval_input_reader.tf_record_input_reader
  

start to Train

cd research/object_detection

python3 model_main.py \
--pipeline_config_path=data/ssd_mobilenet_v1_coco.config \
--model_dir=data/training \
--num_train_steps=60000 \
--num_eval_steps=20 \
--alsologtostderr

Monitoring training

tensorboard --logdir=object_detection/data/training

Convert the checkpoint to frozen graph (.pb file)

cd research/object_detection

python3 export_inference_graph.py \
--input_type=image_tensor \
--pipeline_config_path=data/ssd_mobilenet_v1_coco.config \
--trained_checkpoint_prefix=data/training/model.ckpt-60000 \
--output_directory=data/training

Evaluation

see https://www.youtube.com/watch?v=rprKW1_TYhw&feature=youtu.be