deep learning - What is a good outdoor camera model for a "camera trap" used to collect data for DL training?

Question

I'm looking for "semi-smart" cameras that suit my needs for collecting training images that will later be used to train a DL model. The closest I've found are cameras from Axis Communication, but I'm hoping I can find better (for my needs).

Goal summary: Collecting outdoor images of animals that walk through a certain area (a configurable motion detection region) at high resolution and low/modest frame rates.

Required:

1. Camera housing and robustness suitable for outdoors, freezing temps, wind, precipitation.
2. 10+ frames per second
3. 4k+ resolution
4. Rectilinear view (or barrel correction, but preferably optical)
5. ~12 ft wide field-of-view
6. Working distance of ~30 ft, but hoping for optical zoom, so not a tight constraint
7. Motion detection: Triggers on motion of large, slow objects (configurable), and ideally can capture images from about -1 to +1 seconds, relative to trigger event moment. Other frames are not needed. Most frames will be discarded.
8. Little or no compression artifacts. MJPEG, perhaps. No H.264/5.
9. Remote real-time streaming just to aid in setup (low resolution, low FPS is ok). Not needed during regular image acquisition periods (which will run for days).
10. Price near $1k

Very helpful:

1. IP, PoE
2. Optical zoom, controllable over IP
3. Only "wanted" images are transmitted from camera. High bandwidth (GigE) not needed since most frames will not meet motion detection criteria, and be at modest FPS.
4. Images stored via IP/ethernet to local machine

Helpful:

1. Pan and tilt, controllable over IP
2. Synchronized triggering between 2 or more cameras (e.g., one camera does motion detection, and triggers all the cameras)
3. No special software/app needed. E.g., can configure and control from a web browser.

Hoping some of you can point me in a helpful direction!! Thanks!

Answer 1

等待大神解答