Classification | Object Detection | Pose Estimation |
---|---|---|
![]() |
![]() |
![]() |
- A linux machine, I tested this build on x86_64
- Front Cam or USB Cam connected to your computer with drivers preinstalled
- Optional - EdgeTPU for NN acceleration
Links attached for your reference
- OpenCV
- Upgrade GCC - Upgrade your gcc and g++ to the latest version. Don't worry they are always backward compatible.
$ sudo apt install libgtk2.0-dev
$ sudo apt search libgtk2.0-dev
- Clone the Repository
$ git clone https://github.com/Eashwar93/coral_edgetpu_video_inference.git
$ cd coral_edgetpu_video_inference/scripts
- Upgrade your cmake to the latest version and again don't worry they are backward compatible.
$ bash install_cmake.sh
- Build the repostitory
$ cd ..
$ mkdir build && cd build
$ cmake ..
$ make
$ cd ..
- If you have a Coral USB Accelerator you can run any of the following scripts else skip to Step 2. The first script runs classification on a video stream,the second runs Object Detection and the third runs human pose estimation
$ bash scripts/classification/classify_edgetpu.sh
$ bash scripts/detection/detect_edgetpu.sh
$ bash scripts/pose_estimation/pose_edgetpu_480x640.sh
- The following scripts can be run on computers without any accelerators:
$ bash scripts/classification/classify_cpu.sh
$ bash scripts/detection/detect_cpu.sh
$ bash scripts/pose_estimation/pose_cpu_353x481.sh
I apprently made use of Coral USB Accelerator and below are the results for your reference. Click here to see the GIF Demos.
- A medium article on how to use Google TPU's and why do we need them.
- ROS wrapper to take advantage of the the TPU for Robot Naviagtion.