In this study, we propose a novel hybrid deep learning model that integrates the BP deep neural network, CNN and LSTM. Based on computer vision technology, the model is designed to deal with real-time object detection and recognition challenges. The COCO, PASCAL VOC and KITTI datasets were used to evaluate the model which was compared against SSD MobileNetV3, YOLOv3 and Faster R-CNN. The results show that the hybrid BP-LSTM model achieves superior accuracy (F1-score: 0.916), while at the same time maintaining competitive inference time (35 ms). Object detection and recognition capabilities of the model are greatly enhanced under simulated real-world scenarios, including autonomous driving, medical imaging systems and surveillance installations. Statistical checks show the significance of these improvements. Although computational requirements are slightly higher from it this balance model still made the trade-offs worth it. Such findings imply that the hybrid method proposed here balances accuracy and speed effectively and, if employed, will be able to provide real-time computer vision applications in various realms.