我已经使用MATLAB Coder将MATLAB脚本转换为C。该脚本以uint 8格式的视频作为输入，使用自定义的YOLOv 4 DL网络执行对象检测任务以识别感兴趣的对象，并创建在对象边界内裁剪的新uint 8数组（视频）（参见下面的代码）：

function [I] = Android3(Vrecorded)

%#codegen
% Loads and created pretrained YOLO v4 DL NW detector customized 
% to detect LED array.

detector = coder.loadDeepLearningNetwork("trainedDetector_23022023.mat");

% Read all frames as grayscale images
frames = reshape(Vrecorded(:,:,1,:), size(Vrecorded,1), size(Vrecorded,2), []);

% extrai o número de quadros do vídeo capturado (NoF - Number of Frames)
NoF = size(frames,3);

%Detect objects in an unknown image / video by using the pretrained YOLO v4
FrameForDetection = zeros(size(Vrecorded,1),size(Vrecorded,2),3);
FrameForDetection(:,:,:) = Vrecorded(:,:,:,1);

%object detector
[bboxes,~,~] = detect(detector,FrameForDetection);

% This cell contains the cropped frames containing RoI after YOLO v4 object
% detection algoritm process the frames
croppedFrameVectors = cell(1,1);

sizes = imcrop(frames(:,:,1),bboxes(1,:));

croppedFrames = zeros(size(sizes,1),size(sizes,2),NoF);

for n = 1:1
    croppedFrameVectors{n} = rand(0,0,0);
    for k = 1:NoF
        % crops the frames in each detected object according to bounding
        % boxes results captured on the first frame
        croppedFrames(:,:,k) = imcrop(frames(:,:,k),bboxes(1,:));
    end
    % Insert the cropped frames with bounding boxes within a cell to allow
    % multiple cropped objects within the same variable
    croppedFrameVectors{n} = croppedFrames;
    % Erase variable to crop another object due to bbox different
    % dimensions from one object to another (widht x height)
end

I = croppedFrameVectors{1};

字符串
其中Vrecorded是大小为240 x320 xF的视频矩阵，并且F取决于输入视频而变化。
在将其转换为C并尝试将其调整为JNI语法之后，这是我迄今为止所做的：

#include <jni.h>
/* Include files */
#include "main.h"
#include "Android3.h"
#include "Android3_emxAPI.h"
#include "Android3_terminate.h"
#include "Android3_types.h"
#include "rt_nonfinite.h"
#include <android/log.h>

JNIEXPORT jbyteArray JNICALL
Java_com_example_occ_MainActivity_teste(JNIEnv *env, jobject thiz, jbyteArray Vrecorded) {
    emxArray_real_T *b_I;
    //emxArray_uint8_T *Vrecorded;

    // Call the entry-point function 'Android3'
    emxInitArray_real_T(&b_I, 3);
    Android3(Vrecorded, b_I);

    // Clean up memory
    emxDestroyArray_uint8_T(Vrecorded);
    emxDestroyArray_real_T(b_I);
}

型
其中Android 3（）是从MATLAB代码执行图像处理的函数。我的Java主应用程序是：

package com.example.occ;

import androidx.appcompat.app.AppCompatActivity;

import android.os.Bundle;
import android.widget.TextView;

import com.example.occ.databinding.ActivityMainBinding;

import java.util.Arrays;

public class MainActivity extends AppCompatActivity {

    // Used to load the 'occ' library on application startup.
    static {
        System.loadLibrary("occ");
    }

    private ActivityMainBinding binding;

    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);

        byte[] c = new MainActivity().teste(I);
        System.out.println("Value of matrix B is: " + Arrays.toString(c));

    }

    public native byte[] teste(byte[] I);
}

型
到目前为止，我有两个问题：
1.我将矩阵Vrecorded（uint 8格式）传递给JNI函数的逻辑是否正确（使用byteArray[]）。该矩阵具有3个维度;
1.如果我在编译运行前不知道矩阵的大小，如何将矩阵'I'返回到Java，以便它可以显示在屏幕上**？**