This project implements object detection served over grpc using the Google Edge TPU. These detections are returned from the server in the form of the centroid coordinates of the object's bounding box and its label.
This is part of the radar-ml project.
Clone this directory and cd to it.
Install the required Python modules in requirements.txt and OpenCV per the instructions here.
If you'd like to compile the Protocol Buffers from scratch, use the following command.
$ python3 -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. ./detection_server.protoSee detection_server.proto for the server's interface definitions.
The following illustrates an example client calling the detection server's services.
"""
Example client for detection_server.py.
Copyright (c) 2020 Lindo St. Angel
"""
import grpc
import detection_server_pb2
import detection_server_pb2_grpc
def get_camera_resolution(stub):
request = detection_server_pb2.Empty()
try:
response = stub.GetCameraResolution(request)
print('Camera resolution fetched.')
return response
except grpc.RpcError as err:
print(err.details()) #pylint: disable=no-member
print('{}, {}'.format(err.code().name, err.code().value)) #pylint: disable=no-member
exit(1)
def get_detected_objects(stub):
request = detection_server_pb2.DesiredLabels(labels=['person', 'dog', 'cat'])
try:
response = stub.GetDetectedObjects(request)
print('Detected object(s) fetched.')
return response
except grpc.RpcError as err:
print(err.details()) #pylint: disable=no-member
print('{}, {}'.format(err.code().name, err.code().value)) #pylint: disable=no-member
exit(1)
def run():
with grpc.insecure_channel('localhost:50051') as channel:
stub = detection_server_pb2_grpc.DetectionServerStub(channel)
while True:
res = get_camera_resolution(stub)
print(res)
res = get_detected_objects(stub)
print(res)
if __name__ == '__main__':
run()Everything here is licensed under the MIT license.
For questions or comments about this project please contact the author goruck (Lindo St. Angel) at {lindostangel} AT {gmail} DOT {com}.
This project inspired by and borrows from OpenCV camera examples with Coral.