Custom deep stream pipeline for fish eye stream Processing

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Specifications
**• Hardware Platform : NVIDIA Geforce RTX 4070 TI
**• DeepStream Version: 7.0
**• TensorRT Version: 8.9.4
**• NVIDIA GPU Driver Version : 470
**• Issue Type : questions

For our usecase we are currently exploring we want to input fish eye camera stream using RTSP and split it into 4 different views. We want to perform inference , tracking and analytics seperately in all four substreams. Please suggest a way to do same.

Fisheye camera RTSP stream → Decoding (H.264) → Dewarping → Sub-stream - 1/2/3/4

                Sub-stream-1  ->  AI model inference (object detection)  ->  Object tracker  ->  Analytics                      
                Sub-stream-2  ->  AI model inference (object detection)  ->  Object tracker  ->  Analytics 
                Sub-stream-3  ->  AI model inference (object detection)  ->  Object tracker  ->  Analytics 
                Sub-stream-4  ->  AI model inference (object detection)  ->  Object tracker  ->  Analytics

image (1)
image (1)602×531 378 KB

image
image999×634 87.7 KB

3

Please refer to Gst-nvdewarper — DeepStream documentation 6.4 documentation.

4

@Fiona.Chen Can you point out documention or examples to connect 4 dewrapped surfaces to seperate infer and tracking modules.

5

You may refer to How to separate nvdewarper with 4 surfers to become 4 separate streams - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums