高通第三方算法（单帧数据）添加流程---hal1/hal3
这篇文档写了很久了，当时脑袋打结用的英文，懒得改成中文了，将就看下，如有错误请指正，谢谢
上图是我总结的高通camera架构。下面介绍怎么添加单帧的算法，其实就是对单帧数据进行处理。
Add 3rd algorithm of singleframe：
As we known，camera consists ofthree parts：preview，capture and video。
If we want to Add 3rdalgorithm,we must find out the data callbacked to the HAL that processed by ISP。
callback file：hardware\qcom\camera\QCamera2\HAL\QCamera2HWICallbacks.cpp
Preview/Video callback：
voidQCamera2HardwareInterface::synchronous_stream_cb_routine(
mm_camera_super_buf_t *super_frame,QCameraStream * stream,
void *userdata)
Capture callback:
voidQCamera2HardwareInterface::zsl_channel_cb(mm_camera_super_buf_t*recvd_frame,
void *userdata)
voidQCamera2HardwareInterface::postproc_channel_cb_routine(mm_camera_super_buf_t*recvd_frame, void *userdata)
The key is the structure mm_camera_super_buf_t,next show the struct of it。
typedef struct {
uint32_t camera_handle;
uint32_tch_id;
uint32_t num_bufs;
uint8_t bUnlockAEC;
uint8_t bReadyForPrepareSnapshot;
mm_camera_buf_def_t* bufs[MAX_STREAM_NUM_IN_BUNDLE];
} mm_camera_super_buf_t;
The buffer is inside in struct mm_camera_buf_def;
typedef struct mm_camera_buf_def {
uint32_t stream_id;
cam_stream_type_t stream_type;
cam_stream_buf_type buf_type;
uint32_t buf_idx;
uint8_t is_uv_subsampled;
struct timespec ts;
uint32_t frame_idx;
union {
mm_camera_plane_buf_def_t planes_buf;
mm_camera_user_buf_def_t user_buf;
};
int fd;
void *buffer;//this stored the data of frame,we shouldoperate it。
size_t frame_len;
void *mem_info;
uint32_t flags;
} mm_camera_buf_def_t;
(Samplecode)Add Face Detection Algorithm:
这里面的so文件就是我们的算法库，通过dlopen函数对其进行操作。
在最开始需要定义：typedef int (*CAC_FUNC)(unsigned char *data, int width, int height);这个与算法接口是对应的。
也需要加入头文件：#include “dlfcn.h"
Position：Because realize FaceDetection just need to operate the buffer (one frame)that stored the data inthe forms of YUV420,so we can add Face Detection Algorithm everywhere in the function synchronous_stream_cb_routine。
Add algorithm step by step:
1：Realize FaceDetection。
2：The Face Detection Algorithm depend on opencv lib。Compile it to be a so file (libmy.so) linked with libopencv_java3.so, then it will be private。
3：Push the related lib to the correct dir of your machine。haarcascade_frontalface_alt2.xml---->system/vendor/lib/haarcascade_frontalface_alt2.xml
libopencv_java3.so ---->/system/vendor/lib/libopencv_java3.so
libmy.so ---->/system/vendor/lib/libmy.so
以上是跑的hal1通道，如果跑的hal3通道的话，需要设置hal3 enable。（高通自带的相机就算打开hal3也是跑的hal1，应该是不支持hal3的，需另装apk）
设置命令是：adb shell setprop persist.camera.HAL3.enabled 1
Hal3的回调文件是：Qcamera3Channel.cpp
预览的回调是在：
voidQCamera3ProcessingChannel::streamCbRoutine(mm_camera_super_buf_t *super_frame,
QCamera3Stream *stream)
拍照的回调是在：
voidQCamera3PicChannel::streamCbRoutine(mm_camera_super_buf_t *super_frame,
QCamera3Stream*stream)
如上图所示就是在预览回调函数里面处理buffer，实现预览变灰的效果。