as-you-like / openvino-realtime-vehicle-detection

Implementation of various models for vehicle detection

movidius-rpi

Vehicle detection examples with OpenVINO running on Raspberry Pi 3 Model B (RPi) and Intel® Core™ i7-4790K CPU @ 4.00GHz × 8 (desktop)

Requirements

• Raspberry Pi 3 Model B
• Movidius Neural Compute Stick 2 (not tested yet on NCS)
• Webcam
• Display (Optional)

Scripts

• realtime_vehicledetection.py (model: vehicle-detection-adas-0002
• realtime_objectdetection.py (model: mobilenet-ssd)
• realtime_objectdetection_and_tracking.py (model: mobilenet-ssd)

Installation

To install the Intel® Distribution of OpenVINO™ Toolkit for Raspbian, follow these instructions

Model conversion to IR

mobilenet-ssd from model zoo has been transformed to IR with model optimizer with following command:

python3 ~/intel/computer_vision_sdk/deployment_tools/model_optimizer/mo.py --input_model ~/workspace/open_model_zoo/model_downloader/object_detection/common/mobilenet-ssd/caffe/mobilenet-ssd.caffemodel --input_proto ~/workspace/open_model_zoo/model_downloader/object_detection/common/mobilenet-ssd/caffe/mobilenet-ssd.prototxt --data_type FP16

Change path according to your installation.

Model Performance

task (model)	Device	FPS
Vehicle Detection (vehicle-detection-adas-0002)	desktop	6.46
Object Detection (mobilenet-ssd)	desktop	18.31
Object Detection And Tracking with default configuration (mobilenet-ssd)	desktop	30.27
Vehicle Detection (vehicle-detection-adas-0002)	RPi	-
Object Detection (mobilenet-ssd)	RPi	6.90
Object Detection And Tracking with default configuration (mobilenet-ssd)	RPi	10.17

References

Thanks to Adrian Rosenbrock for the following helpful tutorials:

• Real-time object detection on the Raspberry Pi with the Movidius NCS
• OpenCV People Counter