robot cameraman with pan-tilt head
  • Hi,
    I'd like to build a robot cameraman similar to Soloshot: https://soloshot.com/
    Soloshot is a pan tilt zoom (PTZ) camera that is mounted on a tripod and films an auto tracked target (e.g. surfer, skier or drone).

    My approach is to control a PTZ camera (mounted on a tripod) using a Raspberry PI (or something similar). There are existing (surveillance) PTZ cameras, but I have not found a camera that has an large enough optical zoom (at least 50x) at an affordable price. Hence, I tend to use a regular camera (DSLR, camcorder, etc.) that is controllable (e.g. using the Sony Camera Remote API) and mounted on a motorized pan-tilt head. I have not found a pan-tilt head that is fast enough for auto tracking (and not only for slow time lapse). A gimbal should also fit my requirements, but I don't want to build that gimbal myself if I can buy a finished gimbal with a documented API or SDK to control it. I'm new to BaseCam products, but it seems that they have developers in mind. Sadly, they only offer a gimbal for smartphones: "BaseCam Handy – 3-Axis Smartphone Stabilizer".

    Does anyone know a gimbal suitable for camcorders or DSLRs that has a documented API to control it (using a Raspberry PI)?

  • I am familiar with what your trying to accomplish however robot needs intelligence this can be derived from image recondition meany such cameras and API already on the market like PIXY & others. regarding basecam electronics BGC gimbals you can buy a DYI kit for LARGE DSLR you assemble your self or ready to use one link below cheaper models FYI there is a learning curve involved!


  • all custom gimbal can be mounted on top or on bottom on tripod or robot or RC truck , copter , airplane up to you what you want to do there are accessory to mount any situation if your not handy! they all support basecam electronics come with BGC Controller.
  • I ended up using a Ikan Beholder DS2-A gimbal controlled by a Raspberry PI (connected by USB cable) that uses a Google Coral USB Accelerator do run (Tensorflow Lite) object detection in real time on the live view image of a Panasonic camera received using the (reverse engineered) WiFi API (also used to control zoom).

    This works quite good so far. The latency of the communication between Raspberry PI and camera is a technical restriction. It would be better to use a cable. Maybe SONY's LANC or Multiport cable can be used, but those protocols are not officially documented and have to be reverse engineered. I might look into this later.

    Another issue is the lower accuracy of Tensorflow Lite compared to Tensorflow. However, AFAIK there is no faster hardware (at the moment) to achieve more accurate object detection (non lite Tensorflow) that is as mobile (energy efficient) as the Google Coral USB accelerator. Computer vision capabilities of the Raspberry PI (or similar devices) should be at least as limited as the AI approach.

    In general, the possibilities (available hardware and algorithms) can be utilized even better. Nevertheless, the computation speed is a major limiting factor. I am curious to see how far I can exploit the possibilities. I appreciate any improvement suggestions and ideas I get ;-)

    @AOPEN3434 Thanks for your input so far :-)