Building Real-Time RTSP Video Stream Pipelines with Python, OpenCV, and InsightFace

Processing live RTSP feeds for edge AI applications requires careful thread management to prevent frame lag and memory leaks. If you process every single frame synchronously, your pipeline will quickly fall behind the buffer.

To solve this, I implemented a multi-threaded frame grabber pattern using Python's threading module combined with InsightFace for bounding box extraction and embedding generation:

~/ code

import cv2, threading, queue
~/ code
class RTSPStream:
    def __init__(self, src):
        self.stream = cv2.VideoCapture(src)
        self.q = queue.Queue(maxsize=2)
        self.stopped = False
        
    def start(self):
        threading.Thread(target=self.update, daemon=True).start()
        return self
        
    def update(self):
        while not self.stopped:
            if not self.stream.isOpened():
                break
            grabbed, frame = self.stream.read()
            if not grabbed:
                self.stop()
                break
            if self.q.full():
                try: self.q.get_nowait()
                except queue.Empty: pass
            self.q.put(frame)
            
    def read(self):
        return self.q.get()
        
    def stop(self):
        self.stopped = True
        self.stream.release()</code></pre><br/>
This keeps latency minimal by dropping stale frames when inference bottlenecks occur. What strategies are you using for edge video pipelines?