Video Systems and Video Analysis

The client approached us with a project to enhance their video-based people tracking system. They required bounding box detection and ID tracking for an already operational neural network. The initial data provided by the client consisted of 100 hours of video

The task at hand involved addressing instances where individuals moved beyond the camera’s field of view, causing the existing neural network to lose track. Our objective was to maintain continuity in tracking a person’s movement, both when leaving and returning to the camera’s coverage area. The client reached out to us due to a lack of in-house taggers capable of handling such voluminous and complex tasks

We deployed ID Track teams in a two-stage process: the first team, comprised of 30 individuals, focused on detection and track creation. Subsequently, a second team of 6 individuals, with overlapping responsibilities, assigned a unified ID to distinct tracks, effectively addressing the issue of losing track when a person moved out of the camera’s range

• Enhanced neural network metrics resulted in a 14% improvement in detecting individuals
• The client now possesses a robust technology for indoor people tracking, even in scenarios involving interruptions in movement (outside and within the camera coverage area)
• We extended our collaboration to long-term projects involving marking tracks of individuals in forest park settings