Applied computer vision projects
Selected projects · Robotics · Tracking · Vision + LLM
A collection of applied vision prototypes spanning augmented interaction, vision-guided robotics, road-scene detection, human pose tracking and real-time electronics control through a vision + LLM chatbot.
Hand tracking and physics-based interaction
A pinch gesture controls a virtual ball
Python prototype using MediaPipe and OpenCV: pinch to grab and move the ball, then release to see momentum, friction and collisions.
More computer vision projects
Seven additional project cases are presented across eight clips, covering visual tracking, robotics, product OCR and conversational control.
Vision-guided robot control
A 3D robotic arm appears alongside a camera view with visual detections. The demo illustrates a computer-vision-assisted robotics workflow in a simulation/prototype setting.
Computer vision · Robotics · Simulation
Road-scene object detection
A road video is annotated with detections such as vehicles and people. The overlay demonstrates object recognition and tracking in a moving-camera scene.
Object detection · Video analytics · Tracking
Multi-person detection and pose tracking
People are detected and assigned tracking IDs while body keypoints are overlaid across the scene. This demonstrates multi-person tracking and pose estimation.
Person detection · Pose estimation · Tracking
Hand and object tracking
Hand landmarks and a detected object are displayed together, demonstrating simultaneous gesture tracking and object localization in a live camera view.
Hand landmarks · Object detection · Real-time tracking
Real-time electronics control with a vision + LLM chatbot
Voice/text chatbot commands
Live PTZ movement and command log
A voice/text chatbot sends natural-language commands to a PTZ camera. The companion capture shows directional movements, return to a saved home position and command logging. Together, the clips show conversational commands and the live hardware response.
Ollama · LLM · Voice/text chatbot · PTZ camera · Electronics control
Product identification with proportional-region OCR
A learned vision model identifies the bottle package. The application defines an OCR region proportionally within the package geometry to read the product name relative to its position. The highlighted region follows the package as it moves.
Learned object detection · Proportional OCR region · Motion tracking
User-adapted real-time product OCR
A user-configured learning workflow recognizes a target package and reads its label through OCR. The live view shows the selected reading region, a binarized image and recognized text as the package changes position.
User-adapted vision · OCR · ROI and threshold tuning
These clips document prototypes and demonstrations. Descriptions reflect the behavior visible in the supplied videos.
Technologies
Computer vision · YOLO · OpenCV · MediaPipe · OCR · Proportional ROI · Robotics · LLM · Ollama