Self-driving cars and surveillance systems need to track objects in their surroundings at high speed with extreme precision to be able to swiftly react to the situation. RADAR and LIDAR are commonly used techniques for these applications, but they are bulky and expensive. Low-cost alternatives use imaging-based methods and are not suitable for tracking fast-moving objects. These techniques tend to capture an image of the scene, which is then computationally processed to generate the object location. This amounts to several gigabytes of data that need to be captured and processed every minute, which is not ideal. A smarter approach would be to use compressive sensing to capture only the minimum amount of data that is needed to generate the tracking information, as demonstrated by the authors in this paper.

In this work, the sample is illuminated with six predefined patterns from two orthogonal directions and the resultant light is collected by two parallel single-pixel photodiodes. The illumination patterns encode the object’s tracking information into the 12 single pixel measurements removing the need to capture a megapixel image. Thus, a million-fold reduction of the captured data reduced the processing time to less than a microsecond per frame, which is far less than the time required to capture an image by the fastest commercial cameras. Though the proposed method was limited to a single object tracking, this low-cost approach shows a promising future with a thousand-fold speed improvement over current alternatives.

---

---

 Add Comment

You must log in to add comments.