| Proj No. | A1063-251 |
| Title | Temporal Consistency Enhancement for Robust Object Detection under Physical Attacks |
| Summary | In safety-critical applications such as autonomous driving, surveillance, and robotics, object detectors often suffer degraded performance under physical attacks such as occlusion (partial hiding of targets) and low-light conditions (nighttime, shadows). These challenges lead to missing detections, unstable bounding boxes, and frequent ID-switches. Current solutions mainly focus on retraining detectors, which is costly and detector-specific. This project proposes a lightweight post-processing module that enforces temporal consistency in detection results, without modifying the backbone detector. The module leverages temporal filters (Kalman, Exponential Moving Average) and graph-based matching across frames to repair missing or low-confidence boxes, smooth jittery bounding boxes, and interpolate trajectories during short occlusions. The study will start with a rule-based temporal smoothing baseline, then extend it with more advanced temporal consistency mechanisms. Evaluation will be carried out on public low-light/occlusion datasets (e.g., ExDark, COCO-OCC) and custom real-world sequences. Performance will be assessed on detection accuracy (mAP), temporal stability (ID-switches, jitter), and computational efficiency. Key Objectives • Reformulate detection refinement under physical attacks as a temporal consistency problem • Design a lightweight post-processing module using Kalman/EMA filtering and graph matching • Demonstrate robustness improvements compared to raw detector outputs in low-light/occlusion conditions Student Requirements • Familiarity with Python and basic computer vision libraries (OpenCV, PyTorch/TensorFlow) • Basic understanding of object detection and tracking concepts • Interest in robustness evaluation, dataset handling, and applied perception research |
| Supervisor | Prof Su Rong (Loc:S1 > S1 B1B > S1 B1B 59, Ext: +65 67906042) |
| Co-Supervisor | - |
| RI Co-Supervisor | - |
| Lab | Centre for Advanced Robotics Technology Innovation (CARTIN) (Loc: S2.1-B3-01) |
| Single/Group: | Single |
| Area: | Intelligent Systems and Control Engineering |
| ISP/RI/SMP/SCP?: |