Valence: A continuous affective dimension measuring intrinsic positivity versus negativity on a normalised scale (mapped between to in the SDK).
Valence Spike: A rapid, high-magnitude upward or downward inflection in the continuous valence curve occurring within a brief window of consecutive frames.
Average Valence: The mathematical mean of all valid valence samples recorded for a specific subject across their tracked session frames.
Arousal: A continuous affective dimension measuring physiological and cognitive alertness, intensity, or activation (mapped between to in the SDK).
Arousal Spike: A rapid, high-magnitude upward or downward inflection in the continuous arousal curve occurring within a brief window of consecutive frames.
Average Arousal: The mathematical mean of all valid arousal values logged for a specific subject across the video timeline.
VA Heat Map: A 2D density representation of the Valence-Arousal model, illustrating the clustering and dispersion of emotional trajectory points over time.
Positive Transitions: A total count of emotional transitions where valence shifted from a negative or neutral state into a positive state.
Positive Shift: An individual instance or event of upward movement in emotional valence transitioning into the positive space.
Negative Transitions: A total count of emotional transitions where valence shifted from a positive or neutral state into a negative state.
Negative Shift: An individual instance or event of downward movement in emotional valence transitioning into the negative space.
Mood Shifts: The cumulative count of how many total transitions occurred across both positive and negative spaces ().
Mood Shift Rate: The percentage of frames where a significant emotional transition occurred, representing the overall volatility of the subject's mood:
Positive Trend: A prolonged session window characterised by net-positive valence trajectory and predominance of elevated affect.
Negative Trend: A prolonged session window characterised by sustained sub-zero valence, reflecting extended periods of difficulty, dissatisfaction, or cognitive friction.
Neutral Trend: A prolonged period where valence and arousal hover tightly around baseline resting thresholds without significant excursions into positive or negative quadrants.
Dominant State: The primary discrete Emotion Zone exhibiting the highest average percentage or cumulative duration across the session for the selected subject.
Emotional Distribution: The percentage breakdown and cumulative duration of session time spent within each discrete Emotion Zone (Neutral, Happy, Sad, Angry, Shocked, Surprised, Horrified, Ecstatic).
Subject 0, Subject 1) dynamically assigned by the SDK to isolate and maintain continuous tracking for individual occupants across video frames.
Identity Reference: The snapshot visual thumbnail extracted from a valid frame and displayed in the Visualiser header, enabling researchers to visually verify which occupant corresponds to a given Subject ID.
Baseline: The subject-specific resting facial morphology reference (landmark mesh and Action Unit zero-points). When the baselining pipeline is active, all subsequent Action Units, Valence-Arousal values, and Emotion Zone probabilities are calibrated relative to this reference frame.
Face Count: The total number of unique, valid human faces detected in a specific frame or registered during the initial discovery pass (num_faces).
Status: The real-time operational state of the video processing queue, SDK, or active subject tracking.
Face Validity: The percentage of recorded video frames where a subject’s facial bounding box and 68-point landmark mesh were successfully detected and verified with acceptable model confidence.
Reliability: An aggregate system quality score reflecting model visibility, tracking continuity, and signal stability across the session.
Environment Quality: An aggregate score outputted as a percentage () calculated across multiple environmental inputs—including whether the video is too blurry, too bright, too dark, or if the face is not properly framed.
Detect Eye Obstructions: A boolean flag (true/false) raised when the Action Units and landmarks around the ocular region are invalid, indicating partial or complete obstruction.
Detect Mouth Obstructions: A boolean flag (true/false) raised when the Action Units and landmarks around the oral region are invalid, indicating obstruction by hands, microphones, masks, or objects.
Visual Voice Activity (VVAD): Visual-only speech detection estimating whether an occupant is actively speaking based on lip dynamics and oral opening/closing cadence, independent of audio signals.
Head Pose: The 3D spatial orientation of the subject's head relative to the camera axis, quantified across three rotational axes:
Yaw Deviations: Lateral rotation (turning head left or right; negative values indicate yaw toward central in-cabin displays or side mirrors).
Pitch Deviations: Vertical tilt (nodding head up or down; downward pitch commonly marks gaze toward consoles or phone interfaces).
Roll Deviations: Lateral tilt (tilting the head side-to-side toward either shoulder).
Extreme Angle Frequency: The proportion or frequency of frames where head rotation exceeds acceptable model tracking angles.
Gaze Vector Projection: 3D directional vectors calculated from pupil and eye landmark geometry estimating the occupant's line of sight into the cabin or testing environment.
Muscle Activation: Numerical intensity scoring corresponding to the contraction of specific facial muscle groups (FACS Action Units).
Facial Muscle Peaks: The highest percentage peak of activation reached by a specific Action Unit (e.g., peak contraction of AU04 Brow Lowerer or AU12 Lip Corner Puller) during the recorded session.
Original FPS: Runs the SDK processing at the native capture framerate of the ingested video container file (commonly or ).
Base FPS: The standardised target framerate for SDK video processing ().
Total Frames: The absolute count of image frames recorded and evaluated across the duration of the media file.
Video Splicing: The operation of selecting a range (start and end timestamps) to synchronously export a trimmed .csv segment and an aligned .mp4 video clip.
SDK Latency: The processing time elapsed between ingesting a frame into the core C++ SDK and receiving the computed output.
Application FPS: The real-time rendering and playback framerate maintained by the desktop frontend application while syncing charts, overlays, and video playback.
Active Camera Source: The designated physical or virtual imaging device used during live analysis.
Active Microphone Source: The designated audio capture device feeding sound input into live analysis.
Light Mode: Visualiser’s high-contrast, light-palette interface theme designed for brighter environments and printed PDF documentation exports.
Dark Mode: Visualiser’s low-emission, dark-palette interface theme designed to reduce visual fatigue during extended video analysis in dim testing environments.
The Blueskeye Visualiser desktop application operates an integrated dual-engine runtime: a native C++ computer vision and neural inference core (BM.BSocial-lib, OpenCV) communicating via a low-overhead Rust FFI layer (src-tauri) with a high-performance React/Vite dashboard.
Because video stream decoding, 68-point facial landmark mesh tracking, Action Unit inference, and real-time temporal charting execute concurrently on the client machine, performance is directly bounded by CPU vector extensions, available unified or system memory, and hardware video decoding throughput.
| Component | Minimum Operational Profile | Recommended Enterprise Profile |
|---|---|---|
| Operating System | macOS 12 Monterey or higher Windows 10 64-bit (Build 19041+) |
macOS 14 Sonoma or higher Windows 11 64-bit |
| Processor (macOS) | Apple Silicon M1 (8-core CPU, 4 performance + 4 efficiency) | Apple Silicon M2 Pro / M3 Pro or higher |
| Processor (Windows) | Intel Core i5 (6th Gen or newer) / AMD Ryzen 3 1200+ (requirement: AVX2 / FMA3 support) |
Intel Core i7 (12th Gen+) / AMD Ryzen 7 (5000 series+) 8+ physical cores |
| Memory (RAM) | 8 GB Unified Memory / System RAM | 16 GB+ Unified Memory / 32 GB System RAM |
| Graphics / Acceleration | Integrated GPU with hardware H.264/AVC decode (Intel HD 520+, Apple M-series GPU) | Dedicated GPU (NVIDIA RTX series with NVDEC) or Apple M-series with unified Neural Engine |
| Disk Storage | 2 GB available SSD space | 5 GB+ high-throughput NVMe SSD |
| Display Resolution | pixels | (Full HD) or higher |
| Capture Interface | USB 2.0 (UVC-compliant, 720p @ 30 FPS) | USB 3.2 / Thunderbolt (1080p @ 60 FPS, low latency) |
The Visualiser platform has been benchmarked and validated under the following standard portable hardware profiles:
| Parameter | Validation Environment Specification |
|---|---|
| System Model | MacBook Air (13-inch, M1, 2020) |
| Operating System | macOS Sequoia / Tahoe (Apple Silicon Native) |
| Processor (SoC) | Apple M1 (8-core CPU, 7/8-core GPU, 16-core Neural Engine) |
| Unified Memory | 16 GB LPDDR4X |
| Binary Architecture | aarch64-apple-darwin |
| Observed Throughput | Real-time analysis at 58–60 FPS (1080p source) |
| Installation Size | ~556 MB |
| Memory Footprint | ~1.4 GB RSS active working set (nominal) |
| FPS Processing SDK Latency | 50-60ms |
| Parameter | Validation Environment Specification |
|---|---|
| System Model | Surface Pro12 |
| Operating System | Window 11 Pro |
| Processor (SoC) | Snapdragon x+ 8 core 3.3ghz |
| Unified Memory | 16 GB LPDDR5 |
| Binary Architecture | arm64 |
| Observed Throughput | Real-time analysis at 55-60 FPS (1080p source) |
| Installation Size | ~514 MB |
| Memory Footprint | ~1.2 GB RSS active working set (nominal) |
| FPS Processing SDK Latency | 30-50ms |
| Parameter | Validation Environment Specification |
|---|---|
| System Model | Lenovo |
| Operating System | Windows 10 |
| Processor (SoC) | i5 1135G7 2.4ghz |
| Unified Memory | 8 GB LP DDR4 |
| Binary Architecture | aarch64 |
| Observed Throughput | Real-time analysis at 30–60 FPS (1080p source) |
| Memory Footprint | ~2 GB RSS active working set (nominal) |