HUSKYLENSGeneral

-IDEA- Husky + Storm32 gimbal = object tracking camera?

userHead Account cancelled 2020-08-05 17:35:24 2751 Views4 Replies
Hi,

I dont have time to do interesting projects so I'll post it here :)

Has anyone tried/succeeded in object tracking camera system? I just got my gimbal-setup working and was thinking that it'd be nice to have a real-time object tracking implemented in it. Husky is so small that it could be sitted next to a camera in gimball and then husky would send the movement orders to a gimball controller (Storm32). This could be used in a standing or mounted gimball so you could do a drone-cam that follows an object. In extended, this could be made so that Husky would also control (directly or indirectly) the drone so that it would control the distance to selected object by sensing its size..

https://www.youtube.com/watch?v=J4Bvrvfg920
https://www.youtube.com/watch?v=k69gAOnJqRQ
2020-09-04 10:56:14 This feature may not be until the end of October, but we try to make it earlier.
Combined with mindplus software, the model can be trained in mindplus software and then deployed to HuskyLens. This is expected to be at the end of October. please wait patiently.
userHeadPic Youyou
2020-08-27 15:31:28 I am eager to testing this new sphere feature. Do you have a rough timeline when we can expect it?
Can we also load in pictures for external pics training of the learning machine ?
userHeadPic fksobel
2020-08-24 15:30:51 As I know, the K210 AI chip supports up to 320*240. May huskylens is hard to detect the ball in different ambient light while you just train them with the current function.
But the new function sphere detection is developing. You could try it in the future.
userHeadPic Youyou
2020-08-15 17:38:31 Hi,

I am also looking for a solution to track a ball in outdoor conditions.

I have used the Huskylens firmware HUSKYLENSWithModelV0.4.7Stable because it offers all menus as described in the wikipage. With the version V0.4.9, I get access to only object detection mode and I can't retrieve position data from the learned object.

The best functioning parameters are those set up by default. However, the camera is only 320x240 if I am not mistaken. It limits the possible detection range in depth (from close to far away) in 1-2 meters approximately.

Is there any possibility to teach the huskylens for a ball classifier to improve its detectability and reduces the effect of lighting dependency?
userHeadPic fksobel