An android app which detects a finger moving left to right or right to left while touching the back camera, and prints that direction on the screen.
1. Capture live camera frame
2. Convert frame from RGBA to HSV
3. Apply two masks to capture red color band in the HSV frame, and do bitwise_or operation to combine them.
4. Divide the combined mask into 2 halves, left and right. Compute difference (left half - right half) of means of pixel intensity values and store them in memory.
5. For a sequence of frames, if the above computed difference moves from positive to negative, the direction is left-to-right. Else it's right-to-left.
Most of the above computation happens in the following 3 methods:
https://stackoverflow.com/questions/35111559/capture-video-frames-with-opencv-in-android https://docs.opencv.org/3.4/db/df6/tutorial_erosion_dilatation.html https://stackoverflow.com/questions/32522989/opencv-better-detection-of-red-color https://docs.opencv.org/3.4/da/d97/tutorial_threshold_inRange.html https://blog.codeonion.com/2016/04/09/show-camera-on-android-app-using-opencv-for-android/ https://stackoverflow.com/questions/10660598/android-camera-preview-orientation-in-portrait-mode https://stackoverflow.com/questions/46015299/how-to-apply-mask-to-live-camera-in-android-opencv https://stackoverflow.com/questions/44413952/gradle-implementation-vs-api-configuration https://stackoverflow.com/questions/17165777/open-the-android-native-camera-using-opencv https://stackoverflow.com/questions/35753393/how-to-add-image-as-mask-in-camera-frame-android-opencv
- A direction label is always displayed once initialized, even when there is no motion
- Sometimes, the direction label might be inaccurate when capturing extremely slow or extremely fast motion
- Memory usage and garbage collection could be better
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent