Download Project (2.3 MB)
We present a 3D anomaly detection technique designed to support various applications in changing environmental conditions. The novelty of the work lies in our approach of creating an illumination invariant system tasked with detecting anomalies in a changing environment. Previous efforts have focused on image enhancement techniques that manipulate the intensity values of the image to create a more controlled and unnatural illumination. Since most applications require detecting anomalies in a scene irrespective of the time of day, (lighting conditions or weather conditions present at the time of the frame capture), image enhancement algorithms fail to suppress the illumination differences enough for Background Model (BM) subtraction to be effective. A more effective anomaly detection technique utilizes the 3D scene reconstruction capabilities of structure from motion to create a 3D background model of the environment. By rotating and computing the projectile of the 3D model, pervious work has been shown to effectively eliminate the background by subtracting the newly capture dataset from the BM projectile leaving only the anomalies within the scene. Although previous techniques have proven to work in some cases, these techniques fail when the illumination significantly changes between the capture of the datasets. Our approach completely eliminates the illumination challenges from the anomaly detection problem. The algorithm is based on our previous work in which we have shown a capability to reconstruct a surrounding environment in near real-time speeds. The algorithm, namely Dense Point-cloud Representation (DPR), allows for a 3D reconstruction of a scene using only a single moving camera. Utilizing the 3D models, we compute the volumetric changes between two reconstructed scenes. We measure the success of our technique by evaluating the detection outputs, false alarm rate and computational expense when comparing the two state of the art anomaly detection techniques.
Primary Advisor's Department
Stander Symposium project
Arts and Humanities | Business | Education | Engineering | Life Sciences | Medicine and Health Sciences | Physical Sciences and Mathematics | Social and Behavioral Sciences
"3D Anomaly Detection using Structure from Motion" (2014). Stander Symposium Projects. 377.
Arts and Humanities Commons, Business Commons, Education Commons, Engineering Commons, Life Sciences Commons, Medicine and Health Sciences Commons, Physical Sciences and Mathematics Commons, Social and Behavioral Sciences Commons