ramakrishnamamidi/object-detection-inference

Object detection inference from rtsp ip camera stream using gstreamer

Object detection inference from IP camera RTSP and video stream using GStreamer HW acceleration and OpenCV

Dependencies

• GStreamer 1.18.5-1 and OpenCV 4.5.1

To Build and Compile

• mkdir build
• cd build
• cmake ..
• make

Running object detection with yolov4

./object-detection-inference --type=yolov4(or yolov4-tiny) --min_confidence=0.6 --link="rtsp://cameraip:port/somelivefeed"  

• Weigths and .cfg files to download inside models folder from https://github.com/AlexeyAB/darknet/releases/tag/yolov4

Running object detection with Mobilenet SSD using Caffe framework

./object-detection-inference --type=mobilenet --min_confidence=CONF_VALUE(for example 0.6) --link="rtsp://cameraip:port/somelivefeed"  

• Caffemodel and Prototxt for deploying(download inside models folder): https://github.com/chuanqi305/MobileNet-SSD

Running with HoG + SVM People Detector

./object-detection-inference --type=svm --link="rtsp://cameraip:port/somelivefeed"

To check all available options:

./object-detection-inference --help

TO DO

• Planning to restore and update inference on tensorflow models from object detection API
• Inference on pytorch models from torchvision object detection API
• Add support for inference with onnxruntime, openvino and tensorrt
• Multi-object Tracking (Sort?)

References

• Using GStreamer to receive a video stream and process it with OpenCV:
  https://stackoverflow.com/questions/10403588/adding-opencv-processing-to-gstreamer-application

• HoG + SVM detector:
  https://github.com/opencv/opencv/blob/master/samples/cpp/peopledetect.cpp

• Train your own HoG detector:
  https://github.com/opencv/opencv/blob/master/samples/cpp/train_HOG.cpp

• Object detection using dnn module:
  https://github.com/opencv/opencv/blob/master/samples/dnn/object_detection.cpp

• Sort algorithm:
  https://github.com/david8862/keras-YOLOv3-model-set/tree/master/tracking/cpp_inference/yoloSort