macOS下PHP调用Python 3.9脚本返回Null结果的问题排查

问题场景 #

在VSCode中编写PHP代码，通过Safari浏览器调用Python 3.9脚本时，仅含print语句的脚本可正常返回结果，但包含人脸识别逻辑的脚本始终返回Null。

相关代码 #

PHP代码

<?php 

$result =  shell_exec("/usr/bin/python3  fr/fr_load.py");
var_dump($result);

?>

Python代码

#!/usr/bin/python3


#import the required libraries
import cv2

import joblib

from sys import exit


from sklearn.feature_extraction.text import CountVectorizer
import pandas as pd




# function to detect face from image
def face_detection(image_to_detect):
    #converting the image to grayscale since its required for eigen and fisher faces
    image_to_detect_gray = cv2.cvtColor(image_to_detect, cv2.COLOR_BGR2GRAY)
    
    # load the pretrained model for face detection
    # haarcascade is recommended for Eigenface  
    face_detection_classifier = cv2.CascadeClassifier('php/fr/models/haarcascade_frontalface_default.xml')

    
    # detect all face locations in the image using classifier
    all_face_locations = face_detection_classifier.detectMultiScale(image_to_detect_gray)
    
    # if no faces are detected
    if (len(all_face_locations) == 0):
        return None, None
    
    #splitting the tuple to get four face positions
    x,y,width,height = all_face_locations[0]
    #[0] cuz we assume we have only one face on our data set
    
    #calculating face coordinates
    face_coordinates = image_to_detect_gray[y:y+width, x:x+height]
    
   
    #for traning and testting all images should be at same size (for egien)
    face_coordinates = cv2.resize(face_coordinates,(500,500)) #we can use any number fit for us
    
    #return the face detected and face location
    return face_coordinates, all_face_locations[0] #all_face_locations[0] beacuse there is only one face in each image

names =[]
names.append("steve jobs")   
names.append("ali alramadan") 
names.append("hillary clinton")


#########load recognition model for later use ###########
face_classifier = cv2.face.EigenFaceRecognizer_create()

face_classifier.read("php/fr/models/our model.yml")

        
######## prediction ##############
#path of the image we want to test and predict 
image_to_classify = cv2.imread("php/fr/dataset/testing/load.jpg")



#make a copy of the image cuz we dont want to ruin the actual image by writing on it
image_to_classify_copy = image_to_classify.copy()

#get the face from the image
face_coordinates_classify, box_locations = face_detection(image_to_classify_copy)  

#if no faces are returned
if face_coordinates_classify is None:
    print("There are no faces in the image to classify")
    exit()
    
#if face is returned then predict the face
name_index, distance = face_classifier.predict(face_coordinates_classify) #if distance is big then there is big different between the train image and test image

name = names[name_index]
distance = abs(distance)

print(name)


#draw bounding box and text for the face detected
(x,y,w,h) = box_locations
cv2.rectangle(image_to_classify,(x,y),(x+w, y+h),(0,255,0),2)
cv2.putText(image_to_classify,name,(x,y-5),cv2.FONT_HERSHEY_PLAIN,2.5,(0,255,0),2)



#show the image in window
cv2.imshow("Prediction "+name, cv2.resize(image_to_classify, (500,500)))
cv2.waitKey(0)
cv2.destroyAllWindows()

解决建议 #

1. 移除GUI阻塞代码 #

Python脚本中的cv2.waitKey(0)会让脚本一直等待键盘输入，导致shell_exec无法获取输出。直接注释或删除以下代码块：

#show the image in window
cv2.imshow("Prediction "+name, cv2.resize(image_to_classify, (500,500)))
cv2.waitKey(0)
cv2.destroyAllWindows()

若需本地可视化测试，仅在直接运行Python脚本时保留这段代码。

2. 替换为绝对路径 #

PHP执行时的工作目录可能与脚本预期不符，导致相对路径无法找到依赖文件。将所有文件路径改为绝对路径，例如：

face_detection_classifier = cv2.CascadeClassifier('/Users/yourname/project/php/fr/models/haarcascade_frontalface_default.xml')
face_classifier.read("/Users/yourname/project/php/fr/models/our model.yml")
image_to_classify = cv2.imread("/Users/yourname/project/php/fr/dataset/testing/load.jpg")

需替换为你项目的真实根路径。

3. 捕获错误输出 #

shell_exec仅捕获标准输出，修改PHP代码将错误输出重定向到标准输出，以便排查脚本报错：

<?php 
$result = shell_exec("/usr/bin/python3 fr/fr_load.py 2>&1");
var_dump($result);
?>

4. 检查权限 #

Web服务器（如Apache/Nginx）的执行用户可能没有访问模型、图片等文件的权限：

• 执行chmod -R 755 /path/to/your/project/php/fr设置文件夹权限
• 确保Web用户属于文件所在用户组

5. 验证脚本独立执行性 #

在终端直接运行Python脚本，确认脚本本身可正常输出：

/usr/bin/python3 /path/to/fr/fr_load.py

若终端运行报错，先修复脚本本身问题（如依赖缺失、模型损坏），再通过PHP调用。

内容的提问来源于stack exchange，提问作者ammr mmar