Below are two object detection solutions I implemented which use traditional Computer Vision techniques such as SIFT or Histogram-of-Oriented-Gradients.
In this example, the task was to find a Coca-Cola bottle in images. The final system used two stages: (i) first an ORB (similar to SIFT) detector was trained to find Coca-Cola logos in images. (ii) The detected logos were then classified as being on a bottle or not using a Deep Neural Network. The full system was integrated into the how-old.net website. For more details see here.
How-old.net input image (left) and detected Coca-Cola bottle (right).
For another project, the task was to detect Giraffes in images. All the images were taken with the giraffe facing right, so that an automatic method similar to fingerprint recognition could use the pattern of the giraffe torso to identify the giraffe. This work is based on an Histogram-of-Oriented Gradients method, and the final model was published as a REST API using Flask. For more details see here.
Giraffe images with torso detection superimposed.