Amlogic S905x Human Detection

Human detection on Amlogic S905x devices using MS COCO 2017 datasets. Datasets were filtered using fiftyone library so that it only contains 'person' class. Model used is SSD MobileNetV2 and YOLOv4-tiny. The detection result can be viewed to website on this repository.

Installation

This guide assumes you have installed armbian to your machine. If you haven't, refer to this repository

Building the Model

Follow this step if you want to train the model from scratch. Otherwise, use the trained model provided on ssd and yolo folder

Downloading datasets

Use fiftyone to download datasets of certain class. Refer to fiftyone_coco.ipynb and adjust the path to your own machine.

Training the model

Model are trained on google colab. Use techzizou's tutorial for MobileNetV2 and [YOLOv4-tiny](https://techzizou.com/train-a-custom-yolov4-tiny-object-detector-using-google(-colab-tutorial-for-beginners/ ). The train configs used to train the model on this repository are this file for MobileNet V2 and [this file] for YOLOv4-tiny.

Running the inference from RTSP stream

Clone or download this repo to your Amlogic S905x device To run human detection inference on s905x devices:

1. Adjust the path to saved model & weights
   • https://github.com/faldeus0092/amlogic-s905x-human-detection/blob/9d5dfaf3e7ea9425c748130037c9a118c9412156/main.py#L22-L24
2. Run the program, example main.py --model yolo --api http://localhost:5000/api/footage --url rtsp://KCKS:majuteru5@10.15.40.48:554/Streaming/Channels/1101 --cctv 2. You can adjust the --model to yolo or mobilenet. --api used for API endpoint /api/footage that this website is hosted on (adjust to your IP accordingly). url used for RTSP URL. --cctv used for CCTV number registered on database.
3. The result can be seen on website (adjust the IP according to your host) http://localhost:5000/video_feed/[cctv_id]

To Do

Improving the model performance by using this repository