SMART TRAFFIC CAMERAS
In this case study, we develop a software application that performs object recognition and tracking in video footages. This technique is also known as “computer vision”. Its key functionalities include:
- Receiving video streams from IP cameras, detecting vehicles and tracking their movement to judge if they violate traffic laws (e.g. red light crossing)
- Automatic number plate recognition (ANPR): uses optical character recognition on camera images to automatically read vehicle registration plates.
- Video storage and retrieval: video from RTSP streams (IP camera) will be recorded and/or re-streamed to other clients over protocols HLS, MPEG-DASH… It can also include a transcoding process (decoding/scaling/encoding).
Key techniques implemented in this system are image processing, machine learning, and video recording/streaming:
- Algorithms: Object Detection (Background Subtraction, HAAR, HOG, CNN), Multiple Objects Tracking (Hungarian, Kalman Filter, Optical Flow), Object Recognition/ Classification (SVM, Neural Networks)...
- Libraries/Frameworks: FFmpeg, OpenCV, Tesseract-Ocr, Dlib, Caffe, TensorFlow…
- Video communication protocols: RTSP, ONVIF, HLS, MPEG-DASH, WebRTC…
Beside traffic management, these techniques can also be employed on other application domains, such as unmanned vehicles, smart security camera, people recognition, etc.
Picture. Vehicle tracking and license plate recognition