DD-VAR / object_detection_with_tensorflow

object detection app based on tensorflow api

Target

• detect object in a video or from camera.

Current Status

on my nvidia 1060 (6GB RAM ), the speed is ~20 frame per second, but gpu utilization is below 30%, I will use multiprocessing to speed up.

Image1:

Image2:

Project Structure

From google

object label and detecion api

data

label data

object_detection

object_detection API implemented by google

My module

myutil

• downloadutil.py: helper function for downloading model
• fps_measure.py: utility to measure fps
• queue_seq.py: a helper class to make sure the frame after processed by multi-process is in order.

object_detection_tf_multiprocessing.py

main module for detecting object

Usage

run following command

python object_detection_tf_multiprocessing.py -v /home/scott/Videos/S11E03.mp4 -p 2

• you need replace the video file with your choice
• -p determine how many image detection process you want to run. For now, each detection process will comsune 1.3 GB system RAM, ~1G Video RAM, 30% GPU(GT1060), you need decide this according to your hardware.