Our conferencing systems are intricately linked with camera systems, notably pan-tilt-zoom (PTZ) cameras. These cameras are controlled via metadata from our conference system, such as identifying the last active speaker. Our operations span various market segments which typically host two types of meetings: formal and informal.
Formal Meetings: In these settings, participants are accustomed to the conferencing system and typically remain in fixed positions. This predictability means the metadata, like microphone toggling, is reliable. However, creating presets for camera positions remains a challenge. We have a proof of concept (Po C) that automates this process using basic video processing for one camera type. This Po C could be enhanced with Convolutional Neural Networks (CNNs) trained on diverse unit images to improve automation across various camera types.
Informal Meetings: Here, participants may not consistently use the system as intended, often forgetting to activate their microphones or moving around the room. Challenges include:

• Speaker Tracking: We have developed a Po C using multiple machine learning models to identify active speakers, which can be further integrated and optimized within our software, considering hardware limitations.
• 360 Camera Integration: Implementing a 360 camera in the room's center to dynamically focus on the speaker.

Combined Challenges: In environments with multiple participants, camera angles can be restrictive, often capturing suboptimal views of speakers. There is potential for developing algorithms to optimize speaker representation by intelligently switching between multiple camera feeds. Currently, no Po C exists for this enhancement.
Candidates interested in these technological advancements in conferencing systems are invited to verify the availability of these topics and register their interest at Televic Careers. Please confirm the topic's availability before applying.