Blue Iris Version 6.0.1.24 Now Supports Custom Models

When comparing
YOLOv10 vs. YOLOv8 for CCTV footage, the choice depends on whether your priority is raw speed and efficiency (YOLOv10) or high detection accuracy, particularly for small or distant objects (YOLOv8).
YOLOv10 is faster and more efficient due to its NMS-free (Non-Maximum Suppression) design, making it ideal for edge devices and real-time, low-latency applications, while YOLOv8 is a more mature and versatile, robust model.

Summary Comparison for CCTV

Feature YOLOv8YOLOv10
Best ForReliability, Small Object DetectionReal-time Edge Deployment, High FPS
NMSUses NMS (Post-processing)NMS-free (End-to-End)
Small ObjectsSuperior performanceCan struggle at far distances
SpeedVery FastGenerally faster (lower latency)
EcosystemVery Mature (Robust)Newer (Research-focused)

Key Takeaways for CCTV Surveillance
  • YOLOv10 (End-to-End Efficiency): Because it eliminates the need for Non-Maximum Suppression (NMS), YOLOv10 is often faster at inference, which is a major advantage for real-time monitoring on resource-constrained hardware like NVIDIA Jetson boards. It is designed for maximum efficiency.
  • YOLOv8 (Robustness & Accuracy): For typical security, where cameras are high up and objects (people, vehicles) are small or far away, YOLOv8 generally performs better. It is more reliable at detecting smaller objects in crowded or complex scenes.
  • Vehicle/Object Detection: While YOLOv10 performs well in vehicle detection, some tests show that YOLOv8 has a slight advantage in car detection, while YOLOv10 might be better at detecting smaller, more complex vehicles like bicycles.
  • Low Light/Blurry Conditions: In cases of blurry video or low-light, YOLOv11 and YOLOv8 are often found to be more stable than YOLOv10.
Recommendation
  • Use YOLOv8 if your main concern is detecting every possible object (high recall), including small, distant, or partially occluded objects, or if you need to use additional features like pose estimation or segmentation.
  • Use YOLOv10 if your primary constraint is limited hardware capacity and you need the lowest possible latency and highest FPS, and if the objects are generally closer to the camera.
Note: As of late 2024/2025, newer models like YOLOv11 or YOLO26 have been released, often combining the speed of v10 with the accuracy of v8.
 
  • Like
Reactions: prsmith777, Skinny1, jrbeddow and 3 others
I agree with the helpful comparison between the YOLOv8 and YOLOv10 for the most part, except in my case YOLOv10 seems to run slightly slower (on an Intel i5-8500 with integrated graphics 630). Distant/small object detection seems a bit worse with v10, often taking longer to detect until they (cars/people) got closer in.
My 2 cents...
 
  • Like
Reactions: Tinman
Haven't had a chance to set it up to do anything useful, but I upgraded to 6.0.1.30, and used one of the YOLO models from the modelpack, and was able to get a sensible analysis (albeit very slow on CPU, in the region of 10,000msec. Perhaps I will try on GPU).
 
You must log in or register to reply here.
Top Bottom