Meskipun peraturan keselamatan kerja di sektor konstruksi sudah jelas, kenyataan di lapangan menunjukkan bahwa banyak pekerja masih enggan menggunakan alat pelindung diri (apd). hal ini disebabkan oleh berbagai faktor, seperti kurangnya kesadaran, tekanan kerja, dan kurangnya fasilitas pendukung. akibatnya, risiko kecelakaan kerja yang serius seperti jatuh dari ketinggian atau tertimpa benda berat menjadi semakin tinggi. pendekatan konvensional dalam menangani pelanggaran keselamatan dan kesehatan kerja (k3), seperti teguran lisan atau pemantauan statis melalui cctv, terbukti kurang efektif dalam deteksi dini dan pencegahan pelanggaran. metode-metode ini juga tidak sejalan dengan perkembangan teknologi saat ini. oleh karena itu penelitian ini mengusulkan sebuah sistem deteksi otomatis penggunaan apd pada area konstruksi menggunakan teknik deep learning yolov8. model yolov8 dilatih menggunakan 3.569 dataset sekunder dengan 100 epoch pelatihan dan pembagian 60% untuk training, 20% validasi, serta 20% testing. berdasarkan pengujian pada 90 frame data testing berupa citra real time, model yolov8 menunjukkan kinerja yang bagus dalam mendeteksi 8 kelas penggunaan apd. dengan nilai rata-rata precision, recall, dan f-measure masingmasing sebesar 0,932, 0,806, dan 0,860. hasil ini menunjukkan bahwa sistem mampu mengklasifikasikan objek yang terdeteksi sebagai penggunaan apd dengan tingkat akurasi yang baik. namun, nilai recall yang masih di bawah 1 mengindikasikan adanya beberapa objek, khususnya kelas "tidak pakai kacamata" dan "tidak pakai sepatu", yang gagal terdeteksi. nilai f-measure sebesar 0,860 menunjukkan adanya keseimbangan yang baik antara precision dan recall. kata kunci : alat pelindung diri (apd), computer vision,deep learning, yolo

Despite the clarity of the regulations pertaining to occupational safety in the construction sector, empirical evidence suggests that many workers are reluctant to utilize personal protective equipment. this reluctance can be attributed to various factors, including a lack of awareness, work pressure, and inadequate support facilities. consequently, the risk of severe work accidents, such as falling from heights or being crushed by heavy objects, is elevated. conventional approaches to addressing occupational safety and health violations, such as verbal reprimands or static monitoring through cctv, have demonstrated limited efficacy in the early detection and prevention of violations. these methods are also not aligned with current technological advancements. this research proposes an automatic detection system for personal protective equipment usage in construction areas using the yolov8 deep learning technique. the yolov8 model was trained using 3,569 secondary datasets with 100 training epochs and a division of 60% for training, 20% validation, and 20% testing. the efficacy of the yolov8 model in detecting various types of personal protective equipment usage was assessed through a testing process involving 90 frames of real-time images. the model demonstrated a commendable performance, exhibiting an average precision of 0.932, an average recall of 0.806, and an average f-measure of 0.860. these results substantiate the system's capability to accurately classify detected objects as personal protective equipment usage. however, the recall value, which remains below 1, suggests that there are some objects, particularly the "not wearing glasses" and "not wearing shoes" classes, that were not detected. the f-measure value of 0.860 indicates a satisfactory balance between precision and recall. keywords: personal protective equipment, computer vision, deep learning, yolov8