What's the best object detection model for detecting people and vehicles in real-time video streams?

For real-time detection of people and vehicles in video, the best choice is usually a YOLO-family model — especially one of the newer variants like YOLOv8 or YOLOv10.

Best overall pick

• YOLOv8s / YOLOv8m: strong balance of accuracy, speed, and easy deployment
• YOLOv10: very competitive for real-time use, with efficiency improvements
• If you need very high FPS on edge devices: YOLOv8n or YOLOv10n
• If you need higher accuracy and have a stronger GPU: YOLOv8m/l or similar

Why YOLO is a good fit

• Designed for real-time inference
• Good performance on common classes like person, car, truck, bus, motorcycle, bicycle
• Wide support in frameworks like PyTorch, ONNX, TensorRT, OpenVINO, TensorFlow
• Easy to fine-tune on your own traffic/camera data

If you want alternatives

• RT-DETR: very good accuracy and more modern architecture, but often a bit heavier/slower than the best YOLO options in practical real-time deployments
• EfficientDet: decent, but generally less popular now for fast video applications
• Faster R-CNN: accurate but usually too slow for real-time streams

Practical recommendation

• CPU / edge device / low latency: YOLOv8n or YOLOv10n
• Balanced production setup: YOLOv8s or YOLOv10s
• Best accuracy while staying real-time on a GPU: YOLOv8m/l or YOLOv10m

Extra tip

If your scene is specific, like:

• highway traffic
• indoor surveillance
• drone footage
• nighttime cameras

then fine-tuning on your own data will usually matter more than the exact model choice.

If you want, I can also recommend:

1. the best model for your hardware (CPU, NVIDIA GPU, Jetson, etc.), or
2. a shortlist with FPS/accuracy tradeoffs.