AgentSkillsCN

video-analysis-workflow

为CMJ和跳深动作提供视频分析指导。在处理运动员视频、调试姿态检测、排查分析失败原因，或运行Kinemotion CLI命令时，可参考本指南。

SKILL.md
--- frontmatter
name: video-analysis-workflow
description: Guides video analysis for CMJ and drop jump. Use when processing athlete videos, debugging pose detection, troubleshooting analysis failures, or running kinemotion CLI commands.
allowed-tools: Bash(uv run kinemotion:*), Read, Glob

Video Analysis Workflow

Analysis Commands

bash
# CMJ analysis
uv run kinemotion cmj-analyze <video> [--output debug.mp4]

# Drop Jump analysis
uv run kinemotion dropjump-analyze <video> [--output debug.mp4]

# Batch processing
uv run kinemotion cmj-analyze videos/*.mp4 --batch --workers 4

Quality Presets

PresetUse CaseTrade-off
fastQuick preview, large batchesLower accuracy
balancedDefault, most use casesGood accuracy/speed
accurateValidation, researchBest accuracy, slower
bash
uv run kinemotion cmj-analyze video.mp4 --quality accurate

Debug Output

Always use --output debug.mp4 when:

  • •Metrics seem incorrect
  • •Troubleshooting pose detection
  • •Validating new videos
  • •Training coaches on video quality

The debug video shows:

  • •Skeleton overlay with joint angles
  • •Phase markers (takeoff, landing, peak)
  • •Real-time metrics display

Camera Angle Recommendations

AngleRecommendationReason
45° obliqueRecommendedBoth legs clearly visible, accurate tracking
90° lateralNot recommendedMediaPipe confuses left/right feet (occlusion)
Front/backNot recommendedDepth ambiguity for sagittal plane motion

Troubleshooting

No Takeoff Detected

  1. •Verify video contains complete jump (before, during, after)
  2. •Check athlete is fully visible throughout
  3. •Try --quality accurate for stricter detection
  4. •Review debug video for landmark quality

Invalid Metrics

  1. •Video may be too short (need full jump cycle)
  2. •Athlete may be partially occluded
  3. •Poor lighting affecting pose detection
  4. •Camera shake causing landmark jitter

Jittery Landmarks

  1. •Check lighting conditions (avoid backlighting)
  2. •Ensure stable camera (tripod recommended)
  3. •Verify athlete clothing contrast with background
  4. •Try --quality accurate for better filtering

Rotation Issues (Mobile Videos)

  • •Mobile videos often have rotation metadata
  • •kinemotion handles this automatically via video_io.py
  • •If issues persist, pre-process with: ffmpeg -i input.mp4 -vf "transpose=1" output.mp4

Video Requirements

RequirementSpecification
Frame rate30+ fps (60+ preferred)
Resolution720p minimum
DurationFull jump cycle visible
LightingEven, front-lit preferred
BackgroundContrasting with athlete
CameraStable, tripod recommended

Output Metrics

CMJ Metrics

  • •jump_height_cm: Calculated from flight time
  • •flight_time_ms: Time in air
  • •countermovement_depth_cm: Lowest point before takeoff
  • •takeoff_velocity_m_s: Velocity at ground leave
  • •triple_extension: Hip, knee, ankle angles at takeoff

Drop Jump Metrics

  • •ground_contact_time_ms: Time on ground after drop
  • •flight_time_ms: Time in air after contact
  • •reactive_strength_index: RSI = flight_time / contact_time
  • •drop_height_cm: Initial drop height (if detectable)

Python API Alternative

python
from kinemotion import process_cmj_video, process_dropjump_video

# CMJ
metrics = process_cmj_video("video.mp4", quality="balanced")

# Drop Jump
metrics = process_dropjump_video("video.mp4", quality="balanced")